Lien Smeesters1, Francesca Venturini2,3 and Anna Grazia Mignani4

1 Department of Applied Physics and Photonics, Vrije Universiteit Brussel and Flanders Make, Brussels Photonics (B-PHOT), Pleinlaan 2, B-1050 Brussels, Belgium

2 ZHAW Zurich University of Applied Sciences, School of Engineering, ZHAW Zurich University of Applied Sciences, 8401 Winterthur, Switzerland

3 TOELT LLC, Research and Development, 8600 Duebendorf, Switzerland

4 CNR—Istituto di Fisica Applicata 'Nello Carrara', 50 019 Sesto Fiorentino (FI), Italy

E-mail: lien.smeesters@vub.be, vent@zhaw.ch and a.g.mignani@ifac.cnr.it

The agrifood industry has been subject to strong digitalization and technological advances during the last decades, including the introduction of Internet of Things (IoT), smart machinery, machine vision inspection systems, and the introduction of precision farming [1, 2]. However, continuous technological developments remain indispensable to tackle the current and future challenges, aiming to deliver a sustainable food and agriculture production, and to offer healthy, nutritious and safe food for all of us. Key challenges include population growth, food waste, food and feed quality, global warming, stopping land degradation and minimizing the use of resources and pesticides. The world population is expected to grow to 10 billion people by 2050, as estimated by the Food and Agriculture Organization (FAO), requiring a dramatic food and feed production increase of 60% [3]. On the other hand, one third of all food produced is currently wasted during production, processing, distribution or at the consumer [4]. In addition, food scandals linked to safety and traceability have been marking headlines for centuries, impacting consumer trust, as reported by the EIT Food Trust Report indicating a consumer confidence of only 55% regarding product safety and 43% for product authenticity [5]. Different food contaminants still sneak into the food chain, among others, pesticide residues, pathogenic microorganisms, and mycotoxins [6]. Food fraud, including mislabeling, dilution, and adulteration is still affecting a wide range of products, among others, olive oil, milk, seafood, coffee and sugar [7]. Finally, in view of mitigating climate change, a transition to sustainable agriculture is indispensable, since agrifood accounts for 70% of the global water use and 24% of the greenhouse gas emissions [8], while climate change is envisioned to strongly challenge the agrifood industry increasing the presence of toxins and affecting the soil and harvest quality.

Photonics technologies play a crucial role in tackling these challenges, by the development of novel optical sensors, imaging systems, smart labels, and lighting [9, 10]. As main advantages, optical spectroscopic sensors and imaging systems offer a non-destructive and chemical-free evaluation, suitable for both individual spot checks and in-line autonomous monitoring. In general, photonics technologies have the potential to impact the whole food supply chain (figure 1), revolutionizing the industry [1, 2]. On the field, drones and agricultural machines (e.g. tractors, plucking robots) supplemented with multi-/hyper-spectral imaging technologies are being deployed for irrigation, fertilizer monitoring, pest and diseases detection. Visible cameras feature shape inspection, and the detection of foreign objects and anomalies, while near-infrared spectroscopy (NIRS) enables measuring nutrients, proteins and sugars. Soil evaluation benefits from distributed fiber Bragg detectors, which offer root and moisture monitoring capabilities. Light detection and ranging (Lidar) is a valuable technology for the mapping of plants. Fluorescence spectroscopy is useful to monitor toxins, amino acids, vitamins, allergens, and pigments, while Raman spectroscopy (RS) enables determining biochemical components like sugars, lipids, water, and proteins. Terahertz spectroscopy is employed, among others, to monitor the leaf moisture content. Handheld and smartphone-connected spectrometers are entering the food chain, with the potential for deployment along the whole supply chain. The global market for photonics for precision agriculture (PA) is currently worth around 4.6 billion euros and is expected to virtually double by 2027 to a value of 9.1 billion euros, corresponding to an annual growth of around 15% [2].

Figure 1 Photonics technologies benefitting the full agrifood supply chain.

Download figure:

Standard image High-resolution image Future trends

Continued monitoring and digitalization of the entire food chain, from farm to fork, is indispensable to enhance sustainability, both environmentally and socially, while offering full transparency and traceability. Photonics technology advances are further required to enable a broader deployment, addressing the cost, size, efficiency and sensitivity, while also considering sample handling, technology useability, robustness, system calibration, seasonal and biological variation. In addition, these technologies are also envisioned to contribute to the EU Soil Deal aiming to manage and safeguard soils for future generations [11], as well as to the Farm to Fork Strategy that is defined within the European Green Deal and aims for sustainable, environmental-friendly food systems [12], and to the United Nations Sustainable Development Goals, and in particularly Goal 2 on Zero Hunger.

Driven by the novel developments, the following key trends can be identified:

1  

Multimodal sensors combining different sensing technologies within a single unit or device. Combining multiple optical sensor technologies enables to extend the sensing capabilities, enabling multi-element detection, while offering an improved sensitivity and accuracy [10]. For example, combination of absorption and light scattering can provide both insight into the starch and moisture content while also giving insights in the firmness of the product, and combination of absorption and fluorescence spectroscopy might enable a multi-mycotoxin sensing covering both fluorescent and non-fluorescent toxins. In addition, sensor fusion can be exploited to improve the calibration, standardization, and robustness of the current sensors.

2  

Miniaturization towards handheld devices and photonic integrated circuits (PICs) [13]. The availability of miniaturized and handheld spectrometer units has been empowering mobile spectroscopy [2]. PICs are driving further miniaturization, offering the potential of adding highly precise sensing functionalities on single low-power and miniaturized chips, giving rise to a cost-effective and scalable technology.

3  

Advanced data processing based on machine learning (ML) and artificial intelligence (AI), benefitting the sensitivity and selectivity within predictive, prescriptive and adaptive processes within the whole food supply chain, including soil monitoring, weed management, disease detection, product sorting and food water management [14–16]. The generation of adequate and complete training sets is of indispensable importance to ensure a robust model, while the processing and sensing technologies are ideally being optimized in synergy. Digital twins (DTs), in combination with AI, might enable the generation of digital plants and animals, helping the prediction algorithm.

4  

Sensor-data fusion and enhanced interoperability of data. Future efforts should be made towards the standardization of photonics data, and a network for data transfer and fusion is requested to ensure its implementation in smart agriculture.

5  

Indoor, controlled farming, including vertical urban farming, driven by new sensor and lighting technologies. The controlled environment enables continuous monitoring of the plant's health, maximizing yield and reducing economic losses. State-of-the-art sensing technologies, making use of hyperspectral imaging (HSI), have indicated a 100× higher crop yield than traditional farming, and using 98% less soil and 95% less water [17], while offering the potential to measure macro-elements such as nitrogen, phosphorus, and potassium. Specialized energy-efficient LED and lighting algorithms can furthermore optimize growth and yield. Additionally, a trend towards the application of quantum dots (QD) for spectroscopy and vertical farming can be observed, optimizing e.g. the sunlight spectrum in greenhouses and contributing to a more efficient light use [18, 19].

6  

The journey of space exploration is expected to expand significantly in the near future. This expansion carries the potential to reshape our understanding of the cosmos while offering tangible benefits for life on our home planet. Challenges arise in meeting the nutritional needs of astronauts and future space colonists in the extreme space conditions, given the objectives of returning humans to the Moon, the long-term exploration of Mars, the growth of space tourism, and the continued operation of the International Space Station. The roadmap outlined here shows photonic technologies as highly suitable for various space food applications. Examples include the use of UV lighting for plant growth systems in space, as well as for water purification. Also, spectroscopic and imaging techniques are instrumental for identifying contaminants and monitoring nutritional content, and sensors are essential to detect spoilage or degradation of food products during storage [20].

Photonics for agriculture and food processing is today still emerging, but represents a fast-growing segment with a compound annual growth rate (CAGR) of 14% [20]. Lighting and UV disinfection are dominant segments, since lighting advancements enhance growth in greenhouses and vertical farms, and UV is crucial for e.g. irrigation and plant immunity. Second, imaging systems are of major importance, which can be used on drones and agriculture robots. Hyper—and multi-spectral imaging systems have been increasingly implemented, as well as thermal camera systems. We expect this will be further boosted by emerging imaging technologies, including polarization multispectral imaging, x-ray imaging, and THz imaging [21]. Additionally, we believe the miniaturization of the spectroscopic sensing technologies towards handheld and pocket-size devices has a huge potential that will revolutionize the agriculture and food industry in the coming years. In general, an increased use of photonics technologies along the full supply chain is envisioned, where photonics sensors are implemented in a wide range of devices and infrastructures, such as tractors, drones, robotic arms and mobile robotic modules, machine vision and inspection systems, and storage and transport. The increased availability of real-time and continuous monitoring tools will benefit an improved food quality and safety for all of us.

Roadmap objectives

The roadmap aims to provide an overview of the state-of-the-art photonics technologies benefitting agrifood applications, including a view of their current challenges and future potential. Chapter 1 discusses image-based monitoring and ML processing, while chapter 2 focuses on spectrometry and spectral sensors. Chapter 3 gives a view of the miniaturization of optical sensors, including handheld devices and PICs. Chapter 4 discusses lighting and light source optimization, both for indoor farming and spectroscopy applications. Each chapter includes both general contributions that provide a general overview of the technology, and case-studies that illustrate the technology for a certain challenge within the agrifood industry. Finally, concluding remarks summarize the key messages of this roadmap.

Stefan Paulus and Anne-Katrin Mahlein

Institute of Sugar Beet Research, Holtenser Landstr. 77, 37079 Göttingen, Germany

E-mail: paulus@ifz-goettingen.de and mahlein@ifz-goettingen.de

Status

With the rise of digital plant phenotyping [22, 23] the use of sensors to measure plant traits in greenhouse screening and experimental field sites has become state of the art. By this, the situation of plants in a specific environment can be assessed and described using explainable parameters in high throughput, objectively and reproducibly. By using machine or deep learning approaches, relevant parameters, such as the reaction to a specific stress can be evaluated qualitatively and quantitatively [24].

Commonly used sensor technologies range from simple RGB sensors to more complex multispectral, hyperspectral, thermal sensors or 3D technologies [25]. For all digital phenotyping technologies, data quality and compromise between spatial and spectral resolution are crucial. To bring sensors into the field, commonly they are mounted on uncrewed ground (UGV) or unmanned aerial vehicles (UAV). These vehicles usually are remotely controlled or follow a predefined route over the field. During the last few years, these systems have greatly improved in terms of the level of autonomous operation for data collection [26].

All these technologies assess an enormous amount of data and adequate technologies are required for interpretation and subsequent decision making. The analysis of the datasets is mainly performed by using state-of-the-art AI models that are based on huge human-annotated datasets [27] (see figure 2). A typical workflow therefore combines (i) implementation of a field experiment to acquire training data (ii) data acquisition (iii) data annotation by experts, (iv) training of the ML and deep learning models with feature importance reporting and finally (vi) explanation and result interpretation using agronomic knowledge [28]. With focus on plant pathology and assessment of plant diseases, significant steps have been made regarding fully automated counting of plants and plant coverage description [29], autonomous rating on single plant and on plot area [30] and integration of data acquisition, interpretation and decision making [31].

Figure 2 The workflow of an agricultural experiment. It includes the agricultural question, the data acquisition, the AI processing pipeline including the human expert's input and the target variables like disease or biomass estimation, digital rating or yield prediction.

Download figure:

Standard image High-resolution image

However, most of the methods developed from research did not made it beyond the proof-of-concept level and need intense reworking to enter a product level which can be used and integrated by practical farmers. Nevertheless, these developments are an impressive demonstration of what can be expected in the field of smart farming and PA. Future development will focus on sensor and analysis in a short term. Long term developments will target online processing of the data in the field as well as integration of these routines into practical agriculture.

Current and future challenges

A distinction is made between short-term and long-term approaches for plant phenotyping. Short-term means a period within five to ten years, while long-term describes a time horizon of more than ten years.

Imagine the scenario, that sensors are moved to and within the field by using unmanned vehicles like drones or robots. In recent years, huge effort has been made to increase the automation level of these vehicles in parallel with the car industry's motivation to produce autonomous cars. The big picture of an agricultural field managed by machines, autonomous flying drones [32] driving or walking robots [33] and the fully and integration of high resolution satellite data is hindered by incompatible regulations and laws as well as from unsolved problems in navigation and communication between actors on the field.

Today, researchers use different types of cameras, each camera type has different advantages and disadvantages. Complete camera-carrier systems are available as well as carriers with interchangeable sensors like RGB, multispectral or thermal cameras. Each system requires, in particular when multimodal data (data from different sources) is collected, internal and external calibration to provide proper and valid data that can be linked to further sensor data and provide georeferencing. While spectral signatures were used to recognize stress symptoms [34] the 3D geometry is described by stereo vision or 3D laser-scanners [35]. Each dataset is interpreted by sophisticated AI models. This adds semantics to the data and aims for the definition of a static or dynamic trait describing the plant status or behavior. AI means commonly deep learning approaches, which are much more difficult to explain, compared to traditional ML methods like SVM or Random Forest. However this explanation is needed when targeting or tailoring the hardware to the application requirements. Finally, and besides all the complex research approaches, application within the field and in agricultural practice are demanded. A common example is the transfer from HSI in the greenhouse using hundreds of scanned channels to a cheaper and more robust multispectral open-field approach with five to six spectral channels [28]. If this transfer is successful, the next step of integrating the AI model into the sensor hardware, called edge AI can be tackled. Another aspect and challenge that needs to be addressed is the issue of data availability and data ownership. Current field experiments collect terabytes of data stored in 'closed silos' with metadata described in heterogeneous defined data formats. This makes it impossible to reuse or even to share the data within the community. Acknowledged standards are missing and we need approaches for data harmonization and sharing on an international level [36]). The current situation slows down transdisciplinary collaborations and the adaption of state-of-the-art algorithms to data of existing experiments.

Long-term challenges will need more time and attention as the user group is more heterogeneous. The biggest challenge is the integration of monitoring routines into practical agriculture as sensor based monitoring at different scales (ground (UGV), air (UAV) and orbit (satellites)), AI data analysis, strengthening the trust in AI solutions, defining evaluation routines for algorithms and standards, adapting rules and laws for the use of robots with a high degree of automation, improving the acceptance of digital methods and finally implementing digital flagship showcases with high radiance into the population (figure 3). Furthermore, the integration of multiple heterogeneous data collection stations for field parameters and microclimate into a (inter-) national forecast system for plant diseases and yield is required. This will lead to AI-supported decision-support systems for application recommendations and if used for an automated documentation system for agriculture. Finally, the main challenge will be to define management decisions and adopt daily routines based on the above-described automated approaches.

Figure 3 Challenges for monitoring crops in the field. While robots, explainable AI, edge AI and data sharing are short-term challenges, the integration of monitoring routines into practical agriculture, the implementation of national prognosis systems and the use of automated monitoring systems are depicted to be long-term goals.

Download figure:

Standard image High-resolution image Advances in science and technology to meet challenges

UGV and UAV are dominant tools in plant science and are used especially on experimental field sites where the area is limited, and a high imaging resolution is required. However, their application in practical agriculture is so far very limited. Nevertheless, during the last years robots have been established especially for row crops with a high cost recovery contribution like sugar beet, salad, and vegetables. These robots are mainly used for weed control but integrate sensor racks and on-board analysis systems for a direct decision-making and application like the control of a spraying nozzle. The extension to other non-row crops, to other plant diseases or even the integration of modern field management strategies such as spot farming are possible [37].

All methods that use online processing on the field require a proper AI model that is capable to interpret the data. The ability to explain the model generally has become the gold standard for building trust and deployment of AI systems [38]. Thus, explainable AI (XAI) developed in recent years describes a suite of ML methods that enable humans to understand, trust and better explain models. Nevertheless, the model interpretability must be considered in combination with constraints and requirements like data- and model explainability, fairness and accountability.

As all AI methods rely on available data, data sharing is essential to encourage a multi-disciplinary cooperation. More and more funding organizations have integrated and made mandatory the FAIR principles (findability, accessibility, interoperability, and reusability) for data management [39]. In addition, the MIAPPE protocol has been published, an approach for Minimum Information About a Plant Phenotyping Experiment [40]. Although these tools exist, data sharing lacks problems like data misuse and missing harmonization, non-existing standards and the risk of eroding properties. Nevertheless, approaches based on blockchain ecosystems have been described in the literature, but a solid implementation is still missing [41].

Edge AI devices that integrate the model directly in the hardware just entered the market and enable the collection of semantic data instead of raw data. Providing sensors with a pre-trained model offers the chance to provide highly customized sensors tailored to the problem [42].

Nevertheless, most field experiments focus on a simple experimental setup to investigate methods for one specific factor or information of interest. We need more trials for data assessment in real world situations with mixed infections and combinations with abiotic stress like nutrient or water deficiencies, which increases the complexity of the problem. This complicates the transfer and integration of methods from research and experimental field sites into the demands of practical agriculture but is closer to the real situation.

Furthermore, local laws and regulations should make it possible to validate new applications and technologies in special test fields with limited scope, even beyond previously tested and permitted scenarios. This includes the utilization of larger payloads for research purposes, the use of innovative application technology, and deployment on prototype carrier platforms. This would decisively strengthen the innovative power and, above all, the contemporary adaptation of regulation.

A further aspect is the availability of climate data for the agricultural fields. In the case of plant diseases, for example, decision making can be supported by a proper microclimate forecast of temperature and humidity of adjacent weather stations. By integration of small microclimate stations distributed across the country and provided vendor-independent by companies and individuals, this disease forecast can be fundamentally enhanced in its quality by a national prognosis system. This also requires, an improved infrastructure and availability of specific technologies in rural areas.

Finally, monitoring and an automated decision-support system can help to automate the farmers documentation obligations. This requires a harmonized field and farm management system similar to ISO-bus which controls the devices of a tractor.

Concluding remarks

This summary shows potential technical fields where a disruptive change in the way sensors, data and analysis models will be used in future can be expected. It has been shown, that either prototypes or at least concepts for the changes have been published or currently entered the marked.

Monitoring crop plants in the field is today a multidisciplinary approach covering technicians, farmers, AI-specialists, and engineers. The shown approaches for data sharing highlighted have perhaps the greatest impact, as data availability will encourage researchers worldwide to work on problems with existing data. The limitation here is not technical, but more a motivation problem. Data harmonization and the definition of obligatory standards is the first step to not just combine different disciplines in a team, but to motivate groups of different disciplines to work together internationally.

This, together with the expected development of new technologies such as sensor fusion, Edge AI, and robots will transform agriculture (see figure 3). Here, the idea of PA, doing the right at the right place at the right time, will be leading for all applications.

Acknowledgments

This study was partially funded by the Deutsche Forschungsgemeinschaft (DFG; German Research Foundation) under Germany's Excellence Strategy—EXC 2070-390732324.

In addition the study is partially funded by the German Federal Ministry of Food and Agriculture (BMEL) on the basis of a resolution of the German Bundestag. The project is sponsored by the Federal Agency for Agriculture and Food (Bundesanstalt für Landwirtschaft und Ernährung, BLE) within the framework of the digitization strategy of the BMEL, call funding code 28DE104A18.

David Perpetuini, Daniela Cardone and Arcangelo Merla

Department of Engineering and Geology, University 'D'Annunzio' of Chieti-Pescara, Pescara, 65127, Italy

E-mail: david.perpetuini@unich.it, d.cardone@unich.it and arcangelo.merla@unich.it

Status

Thermal IR imaging encompasses the collection, analysis, and interpretation of data predominantly obtained within the thermal infrared IR portion of the electromagnetic spectrum, and it could be employed in agriculture both on the field or in a remote sensing configuration. IR remote sensing commonly acknowledged in vegetation research spans from 3 to 14 μm, which is further categorized into the mid-wave infrared (MWIR) range of 3–5 μm and the long-wave infrared (LWIR) range of 8–14 μm. The MWIR sensor detects radiation that is emitted by the Earth as well as radiation that is reflected from the Sun. On the other hand, the LWIR sensor is primarily influenced by the emitted radiation [43]. Extensive research has been conducted on the remote sensing of vegetation in the visible and near-infrared (VNIR) range (0.3–1.0 μm) and shortwave infrared (SWIR) range (1.0–2.5 μm), with a specific focus on the analysis of biochemical and biophysical properties of vegetation [44]. Nevertheless, the spectral data obtained from the VNIR and SWIR domains are insufficient in capturing the complete range of structural and chemical attributes exhibited by plants. To face this issue, hyperspectral and multispectral imaging can be employed. Specifically, HSI provides a detailed spectral analysis across a wide range of wavelengths, enabling the detection of specific biochemical properties of plants [45], whereas multispectral imaging, while less detailed than HSI, strikes a balance by providing sufficient information for various agricultural assessments without the complexity of hyperspectral data, which requires complicated processing for its interpretation [45]. Notably, it should be highlighted that the key absorption qualities of specific vegetative components, such as polysaccharides (e.g. cellulose) and leaf surface attributes (e.g. waxes and hairs), are situated in the thermal infrared domain [46]. The advancement of technology in the field of thermal IR spectrometers and sensors has facilitated the increase in the utilization of thermal IR imaging in the agricultural sector in recent years [47]. This can be attributed to the decreased costs of the equipment and the straightforward operational procedures, which have opened up possibilities for its implementation in various areas of the agricultural and food industries. Furthermore, efforts are currently being made to enhance its compatibility with precision farming practices [48]. The thermal characteristics of plant leaves are influenced by a multifaceted and heterogeneous interior structure, which encompasses a specific quantity of water per unit area. Therefore, the potential for conducting research on individual plants using thermal remote sensing is feasible due to the versatile, accurate, and high-resolution capabilities of infrared thermography [47]. However, the precision of thermal measurements is contingent upon the prevailing environmental circumstances, as these factors have a direct impact on the thermal characteristics of the observed crop. Hence, it is imperative to calibrate pictures based on meteorological conditions in order to facilitate the comparison of image data acquired throughout distinct measurement intervals and growth seasons. Therefore, calibration of images according to weather conditions is necessary for comparison between image data obtained during different measuring periods and growth seasons [47]. Moreover, in order to compare images, the soil water content (SWC) should be estimated [49]. In this perspective, an effective approach based on the use of multivariate statistical analysis, such as partial least squares (PLSs) regression, has been proposed for adjusting thermal measurements based on environmental variables [50]. This technique allows for the development of models that can predict thermal responses under varying conditions, thus enhancing the reliability of thermal imaging in agricultural applications. Thermal remote sensing technology has the potential to be applied across various agricultural materials and processes, encompassing situations where heat is either generated or lost throughout spatial and temporal dimensions [51]. The potential application of thermography in the field of agriculture encompasses various areas such as monitoring nurseries, scheduling irrigation and harvesting, detecting, identifying diseases and pathogens, crop health monitoring, evaluating ripeness, and detecting bruises, as reported in figure 4.

Figure 4 Examples of the most spread applications of thermal IR imaging in agriculture. This technology could help in crop monitoring, ripeness and bruises detection, soil conditions evaluation and irrigation management. This figure is created in Biorender.com.

Download figure:

Standard image High-resolution image Current and future challenges

Thermal IR technology has proven very beneficial in PA due to its ability to non-invasively and efficiently monitor and assess many aspects of crop health and environmental conditions. Specifically, thermal IR imaging allows academics and practitioners to get valuable information about how plants react to stress, improve the way resources are managed, and increase agricultural production. Research has highlighted the significance of thermal imaging in studying the relationship between plants and their environment, making it a widely used technology in the fields of agronomy and environmental sciences [52]. In addition, thermal IR imaging is widely recognized as an essential technique in PA for monitoring crop health, identifying stress, and analyzing the environment [53]. Concerning the evaluation of plant stress, thermal IR imaging has played a crucial role in evaluating plant stress in several scenarios, including those with high salt levels and water scarcity in rice farming [54] and citrus trees in greenhouses [55], showing its ability to evaluate the water condition of plants. Particularly, thermal IR imaging is able to assess plant stress responses by tracking fluctuations in leaf temperature, which indicate changes in plant physiology under different stress scenarios. Research has shown the efficacy of thermal imaging in identifying plant distress caused by reduced rates of photosynthesis and transpiration or scarce water condition [52, 56]. Particularly, thermal imaging allows for the real-time evaluation of plant responses to stress factors such as drought by monitoring leaf temperature and angle, which serve as early-stage markers of plant stress [57, 58]. Importantly, the surface temperature of plants is regarded as a highly responsive indication of stress, frequently anticipating the manifestation of apparent symptoms [59]. By integrating thermal imaging with multispectral imaging, researchers may get a holistic comprehension of the physiological reactions of plants under stress [60].

It is worth to highlight that thermal IR imaging was used to evaluate not only the temperature of the leaves, but also of the soil. Particularly, thermal imaging may be used to evaluate SWC and its distribution, and this technology provides extensive insights for this purpose. For instance, Tian et al developed a thermo-time domain reflectometry (T-TDR) probe capable of monitoring the water content and thermal characteristics of unfrozen soil, as well as measuring the amount of ice in partly frozen soils [61]. Similarly, Qiu & Zhao devised an algorithm using thermal imaging and the three-temperatures model (3 T model) to gauge soil evaporation and soil water status [62]. This technique showcases the capacity of thermal imaging to track the movement of water in soil and the pace at which it evaporates.

Thermal IR imaging has emerged as a helpful technique in assessing seeds in nurseries because of its non-destructive characteristics and capacity to provide important information on seed quality and viability. Research has shown that thermal imaging may be used to observe several factors associated with the health and quality of seeds. These factors include seed viability, the health of transplants, the quality of graft unions, and the identification of physiological diseases [52, 63]. Moreover, thermal imaging methods have been used to diagnose seed viability without causing harm during the process of seed absorption, through image processing and statistical algorithms [64]. This demonstrates the efficiency of thermal imaging in evaluating the health of seeds under controlled environmental settings [65, 66]. To examine the disparities between the coldest and hottest seedlings using thermal imaging, it is essential to take into account the thermal reactions of the seedlings under different environmental circumstances. Prior research has emphasized the importance of thermal imaging in observing how plants react to fluctuations in temperature and stressors, demonstrating its capability to assess identify physiological processes such the closing of stomata and water deficiency stress [67, 68]. Regarding seedlings, the thermal response refers to the impact of temperature on the growth and development of plants. Studies have shown that exposing seedlings to warm temperature conditions during their first development stage might affect their ability to withstand stressors such as oxidative stress and transpirational water usage in the future [69].

Thermal IR imaging has been demonstrated to be a powerful method for identifying diseases in plants because it can record temperature differences linked to pathogen infections. Research has emphasized the capacity of thermal imaging to identify and monitor plant diseases by evaluating temperature fluctuations caused by stress imposed by pathogens such as pathogenic bacteria, fungus, nematodes, and viruses [70, 71]. Moreover, thermal imaging has shown potential in identifying the ripeness and quality of fruits and vegetables in agriculture and the food business [72]. Particularly, thermal IR imaging has been employed to discern various phases of fruit ripeness, particularly in unblemished tomato fruits [73], but also to identify excellence of fruits, as well as identifying injuries in fruits and vegetables [74].

Advances in science and technology to meet challenges

Concerning a spread and proper application of this technology in agriculture, it is worth to highlight that some limitations should be addressed and overcome. For instance, the accuracy of thermal imagery can be affected by environmental factors such as weather conditions. To meet this challenge, it is essential to allow the camera to acclimatize to local weather conditions, and to protect it with a casing in order to reduce air temperature effects on the sensor when mounted on UAV [75]. Moreover, algorithms that can adjust the data for weather influences should be developed. Another relevant issue is related to the data management. In fact, handling the vast amount of data generated by thermal imaging can be challenging, hence data management systems should be implemented exploiting IoT solutions. Concerning the improvements of the sensors and data analysis, it should be highlighted that a good spatial resolution should be provided by IR cameras when they are mounted on an UAV. In this perspective, more sensitive optics should be developed, and algorithms able to generate fine spatial resolution thermal images from coarse spatial resolution [76]. To this aim, also the integration of IR imaging with other techniques could help to overcome this limitation. In fact, combining thermal imaging data with geographic information system (GIS) can help in mapping and monitoring the spatial variability of crop fields, aiding in more precise agricultural planning and management [77]. Moreover, the employment of AI algorithms could aid in enhancing the quality of thermal images, by filtering noise and improving the clarity, which could lead to more accurate analysis [78]. AI could also help automate the data analysis process, saving time and reducing the workload on farmers. Moreover, AI models able to forecast crop yields or identify potential pest infestations before they might produce significant problems could be developed. AI can help in the implementation of PA, where farmers can make data-driven decisions to manage their crops more efficiently, and it can facilitate remote monitoring of crop fields, allowing farmers to detect issues without having to physically inspect the fields, thus saving time and resources. In addition, AI systems can provide real-time alerts for potential issues detected through thermal imagery, helping farmers take timely actions to prevent losses. A possible technological hub integration is proposed in figure 5.

Figure 5 A possible technological hub integration for agriculture management. This framework encompasses the integration of IR imaging, eventually mounted on UAV, with AI and IoT algorithms.

Download figure:

Standard image High-resolution image Concluding remarks

IR thermal imaging has the capability to delineate subtle variations in temperature, providing several advantages in agriculture. The most relevant applications of such a technology are related to detect plant stresses even before physical symptoms manifest, evaluating the condition of the soil, to manage the irrigation and to detect bruises, thereby heralding an era of preventative and proactive agriculture management. Leveraging advancements in sensor technology, coupled with the integration of AI, can potentially morph the current challenges of the employment of thermal IR imaging in agriculture into catalysts for further innovation, paving the way for intelligent agricultural solutions characterized by automation, predictive analytics, and real-time monitoring.

Benjamin Gac1,2, Stephane Perrin3, Denis Trégoat3, Luiz Poffo4 and Antoine Fournier1

1 Arvalis, 45 Voie Romaine, 41240 Beauce la Romaine, France

2 Université de Rennes, CNRS, Institut FOTON—UMR 6082, F-22305 Lannion, France

3 Photonics Bretagne, 4 Rue Louis de Broglie, 22300 Lannion, France

4 XLIM UMR CNRS 7252, University of Limoges, Limoges, France

E-mail: b.gac@arvalis.fr, sperrin@photonics-bretagne.com, dtregoat@photonics-bretagne.com, luiz.poffo@unilim.fr and a.fournier@arvalis.fr

Status

Chlorophyll fluorescence is a phenomenon based on the ability of chlorophyll molecules in the plants and algae to release a part of absorbed energy in form of light emission at specific wavelength. This release of absorbed energy is in direct competition with heat dissipation and with the photosynthesis process. Fluorescence measurements can be classified as passive (e.g. natural sunlight) or active excitation methods. Both passive [79] and active [80, 81] chlorophyll fluorescence measurement methods have been investigated using unmanned aerial, ground and manual systems [82]. By analysing the fluorescence emission light, the physiological status, stress level and environmental interactions of plants can be retrieved at different scales [83].

At the leaf level, chlorophyll fluorescence measurements allow the analysis of the plant photosystem II efficiency (i.e. conversion of absorbed light energy into chemical energy). Key parameters can then be extracted, among them, the effective quantum of the photosynthesis process [83]. Furthermore, these measurements have informed on the ability of plants to manage the absorption of light under various stresses, either biotic or abiotic [84].

At the plant scale, the global health of the whole plant has been estimated through the stress-induced changes in fluorescence patterns [85]. In addition, the photosynthetic capacity and efficiency have been studied at different phenological stages. Moreover, depending on the environmental conditions, the fluorescence response can provide information on the adaptation of plants to their surroundings [86].

At the canopy level, chlorophyll fluorescence has been used to probe the photosynthetic activity (PSA) over a large area and to reveal changes in ecosystem dynamics (e.g. responses to climate change and shifts in plant communities) [79]. Also, it allowed to map stress patterns for assessment of regional environmental conditions as well as potential agricultural or ecological challenges (figure 6).

Figure 6 Representation of scales(top), method (middle) and environmental (bottom) established status (in green), current challenges (in orange) and trends of progress (in red) among the community for a broader use of plant fluorescence. This figure has partly been designed using images from Freepik.com.

Download figure: