Technologies developed in agriculture can help farmers to increase crop yield. However, damage caused by pests and diseases increases losses in crop production. Farmers typically carry out the detection of pests and diseases when the crop is in a severe condition. The early detection of pests and diseases using mobile applications is an alternative approach. Farmers can only send a report to the app by providing information that a particular crop is infected. The area of the infected crop will also be displayed on a map, and the system will be synchronised to inform all the farmers to observe their crop and suggest a method of control to prevent the spread to other neighbouring areas.
Sistem Pakar Identifikasi Hama dan Penyakit Padi (Paddy Pest and Disease Specialist Identification System) is another mobile application that identifies pests and diseases in paddy. The information was displayed in tabular form, including the causal agent, common and scientific names of the pest, a picture of the pest, symptoms, and the control method
[32]. Firstly, the user needs to enter the consultation menu from the app and select which types of consultation the user needs, i.e., disease or pest. Secondly, the user needs to insert their name and answer all the questions generated in the system according to the symptoms that appear on the rice. Finally, the result of the consultation will appear, and the user can print or visualise the output through the application. e-RICE is another mobile application that provides information to farmers about pests and diseases in paddy. It uses a rule-based algorithm to classify rule generation based on the knowledge and information provided by experts in paddy to classify the symptoms given by a farmer for an actual diagnosis. Each evaluation of the disease diagnosis will be reviewed again by the developers, other farmers and agricultural officers
[33].
In Malaysia, researchers from Universiti Kebangsaan Malaysia developed a mobile application named Dr Lada
[34]. This application was used to detect pests and diseases in pepper. Users could diagnose the pest or disease infection from this application by answering questions, which minimised farmers’ dependency on an agricultural officer because they were able to diagnose diseases themselves. Furthermore, a research team from the International Rice Research Institute at the University of Queensland, Australia, the Philippine Rice Research Institute in the Philippines, and the Research Institute for Rice in Indonesia developed a mobile application to diagnose pest and disease infestation in crops by answering questions based on the symptoms that appeared. This application was called Rice Doctor and was used to identify the possible ways for the disease to spread and provide suggestions on how to diagnose/treat and overcome the infection. Furthermore, researchers from the National Rice Research Institute in India also developed a mobile application called riceXpert to provide information on the disease, pest, weed, and other possible causal agents that cause an infestation in paddy fields
[47]. Using those mentioned applications, farmers could transfer the data from the field in a user-friendly way in a shorter timeframe to conduct decision making for pest and disease prevention.
Additionally, the mobile application guides farmers to diagnose the infection and further process the application
[51]. For cassava crops, AdSurv collected the infected crop images and labelled them on the images as evidence
[52]. The disease diagnosis is also used to detect major diseases based on the symptoms that appear on the leaf, such as Cassava Mosaic Disease, Cassava Brown Streak Disease, Cassava bacteria blight, and Cassava green mite. The collection of images is divided into five categories of healthy plants. Another four types are based on each disease because each disease has distinct symptoms on leaves
[53]. Therefore, farmers can use this application to diagnose the infected plant and assess the severity of the infection. Hence, image processing is one of the most common methods to visualize the infected plant using mobile applications.
A mobile application was developed to identify disease in crops, for example, paddy, using fuzzy entropy
[54]. Fuzzy entropy is a system capable of modelling non-statistical imprecision and works well for disease extraction. The result showed that fuzzy entropy has more than 90% accuracy in detecting disease in paddy, except for tungro disease, with an accuracy of only approximately 70%. There are four diseases in paddy identified in the study, namely bacterial leaf blight, tungro disease, brown spot, and leaf blast. The camera captures the infected plant for image preprocessing that involves cropping, converting, and enhancement. For image extraction, fuzzy entropy is used to extract the disease. After that, image classification used a Probabilistic Neural Network to classify the disease. The results are shown in the mobile application. As a consequence, various systems in image processing could be used to extract pests and diseases on plants.
3. Spectral Signature Analysis for Pest and Disease Management
3.1. Spectral Reflectance in Monitoring Plant Health
Reflectance is a measure of electromagnetic energy that bounces back from the surface of a material. It is a wavelength-dependent ratio of reflected incident energy. Leaf reflectance in the visible (400 to 700 nm), near-infrared (NIR, 700 to 1100 nm), and shortwave infrared (SWIR, 1100 to 2500 nm) ranges are influenced by a variety of interactions. These interactions involve radiant energy absorption stimulated by leaf chemistry, light scattering due to the leaf surface and internal cellular structures, and radiant energy absorption caused by leaf water content, proteins, or carbon content
[55].
Numerical knowledge of the canopy size is important for efficient farm management. Precision agriculture applications that seek to estimate this commonly use canopy health maps, i.e., as expressed by leaf area per unit (such as plant or meter of cordon), the leaf area index, or other canopy parameters (vegetation fraction and biomass) as a proxy. To correctly map the spatial variability of such farm features, remote sensing data from satellite, aircraft, or drone platforms is needed
[56]. In the case of vertical shoot-positioned canopies, a substantial proportion of soil (bare or with cover crops) is exposed to nadir-viewing remote sensing from the inter-row area. Surface reflectance is subject to fluctuations caused by the canopy structure and its illumination at suggested spatial resolutions, equivalent to plant or row spacings
[57].
Green vegetation of spectral signature feature basins in the visible range of the spectrum has shown pigmentation in plant tissues. Chlorophyll is the major photosynthetic pigment in green vegetation, and it is significantly absorbed in the chlorophyll absorption spectral bands at red (670 nm) and blue (450 nm). When a plant is pressured to the extent that chlorophyll growth is decreased, the amount of reflectance in red (670 nm) regions increases
[58]. Water’s spectral response has different substances characteristic of light absorption NIR and beyond. The suspended sediments and increase in chlorophyll levels are two common elements influencing the spectral response of water. In each situation, the spectral response will change to indicate the presence of suspended sediments or algae in the water
[59].
Detection anomalies in the photosynthetic parameters are crucial in remote sensing approaches. Changes in pigment, nutrients, cell structure, water intake, chemical concentrations, and gas exchange are subsequently displayed in the reflectance characteristics of the leaf or canopy
[60]. The anomalous behaviour is then attributed to abiotic or abiotic stress. Indirectly, the observed and measured changes in spectral reflectance are related to plant stressors
[61]. The data collected by the sensors are often compared using one of the various vegetation indices and subjected to extensive data analysis to be categorised as healthy or unhealthy and between different types of pests and diseases. Every extra step introduces uncertainty into the technique.
As most of the previous studies are specific to the combination of the crop and a pest or disease and relate to external factors, the findings in the literature are non-uniform and it is challenging to compare them quantitatively. As an example, extrinsic factors such as the leaf internal structure, surface features, and water content could influence the pigment absorption of plants. Hence, no single wavelength is associated with a single pigment concentration
[62]. Due to the failures of this method, researchers have turned to correlation analyses to establish unique pathogen-specific spectrum signatures, such as a spectral index and ratio with discriminant analyses
[63][64], but they do not provide conclusive optimal spectral signatures. However, the same findings indicate that the sensitivity of particular spectral areas with significant absorption corresponds to abiotic and biotic factors such as pigmentation
[65].
Figure 3 depicts a framework of plant health monitoring employing spectral signature analysis.
Figure 3. Steps in spectral signature analysis for plant health monitoring.
The use of spectral signatures for pest and plant diseases in the parametric analysis is limited. Non-parametric techniques such as Principal Component Analysis, Cluster Analysis, Support Vector Machines, Partial Least-Squares, and Artificial Neural Networks (ANNs) have been widely adopted by researchers
[66][67][68]. For example, in general terms, PCA is one kind of feature extraction method that helps to find the highest contribution of points, and the highest contribution of points can be identified through the highest eigenvalues with principal component during PCA analysis. Therefore, the lowest contributions amongst those points can be omitted and only the points with the highest contribution are selected for further processing/analysis. Generally, a comparison between thermal, fluorescence, and hyperspectral imaging supports a multi-sensor data fusion method to measure plant health
[69]. A comprehensive study
[70] on head blight on wheat highlighted each system’s main benefits and drawbacks and subsequently studied the individual sensor combinations. Using IR in the 7.5–12 m wavelength region, thermography-based sensors showed temperature differences between crops influenced by biotic and abiotic stresses. While chlorophyll fluorescence-based approaches in the visible spectrum are widely utilised, they are inhibited by the need for dark adaptation to minimise the effect of sunlight on the measurement. Hence, spectral reflectance could be used in monitoring plant health.
3.2. Spectral Signature of Pest and Diseases in the Crop Field
Both hyperspectral imaging and non-imaging sensors are effective techniques for detecting changes in plant health
[71]. Changes in reflectance are caused by plant tissue’s biophysical and biochemical properties. Plant diseases can alter tissue colour and leaf shape, transpiration rate, crop canopy morphology and density, and the interaction of solar radiation with plants
[72]. They cause changes in the optical characteristics of leaf tissue. As there are changes in pigmentation, the hypersensitive response, and cell wall deterioration, leaf reflectance is thus susceptible to plant stress
[73]. Pest and disease-specific symptoms, such as the succession of chlorotic and necrotic tissue with different optical properties and composition, as well as typical fungal structures such as powdery mildews, rusts, and downy mildews, may be identifiable.
When plants are subjected to infections that cause chlorotic and necrotic symptoms, the composition and content of leaf pigments change. The type of host–pathogen interaction influences the pattern of responses and the degree of up- and down-regulation of physiological systems. Necrotrophs rapidly kill plant cells and then feed on the nutrients generated by the dead tissue, whereas biotroph pathogens create haustoria to take nutrients from living cells
[74]. Because the characteristics of the symptoms vary, different wavebands may be appropriate for detecting various diseases. Using sensing techniques, identifying a disease, its discrimination from other diseases, and abiotic stressors is still a challenge in vegetation monitoring. The interpretation of spectral reflectance data without knowledge of the spectral characteristics of leaves and typical symptoms is impossible at present. The highest findings of disease detection were found in the visible and NIR ranges of the spectrum. For example, reflectance spectroscopy was employed to identify the wilt induced by the vascular fungus Fusarium oxysporum from that caused by drought in tomatoes
[75].
Differences in spectra, ratios, or derivations can be used to distinguish changes in spectral reflection and differences in spectral signatures
[76]. This method can compare the spectra of healthy and unhealthy plants. Meng et al.
[77] discovered various important regions of different spectra between healthy plants and plants infected with Cercospora leaf spot, powdery mildew, and sugar beet rust. Based on an understanding of reflectance properties, spectral algorithms for remote sensing of vegetation have been created, which use specific wavelengths of spectral signatures. They are linked to various biochemical and biophysical plant factors that indicate plant health. Spectral vegetation indices are frequently used to monitor, analyse, and map temporal and spatial variation in vegetation.
Disease symptoms can be seen at specific wavelengths and might include any number of changes in the plant’s colour, shape, or functioning as it responds to disease. The disease symptoms vary depending on the pathogen and include leaf spots, chlorosis, necrosis, wilting, or overgrowth. Plant stress other than diseases can activate protective mechanisms that result in suboptimal development, chlorophyll loss, or changes in surface temperatures
[78]. These changes cause noticeable modifications in the spectral signature compared to a healthy plant and may be detected using several approaches
[79].
Based on
Figure 4, visible light can be applied to evaluate variations in the colour and morphology of infected plant tissue. Changes in water content, leaf thickness, and photosynthetic efficiency could be detected using infrared and short-wave infrared, whereby long-wave infrared can be used to monitor plant surface temperatures. Multiple images are captured using hyperspectral sensors over its wavelength range of 300–2500 nm. Furthermore, imaging devices measure the absorption, transmission, and reflectance of input electromagnetic radiation interacting with the plant surface. Compared to healthy tissue, infected plant tissue generally has a lower reflectivity. Image analysis algorithms determine the contrast between diseased and non-diseased leaf areas
[31].
Figure 4. Techniques for high-throughput phenotyping of plants and diseases.
A great deal of research has shown that textural and phrenological differences can also be considered as a viable technique using hyperspectral data for the remote identification of invasive plants such as the Nile rose or water hyacinth (
Eichhornia crassipes). In contrast to previous broadband multispectral sensors, new-generation sensors such as Sentinel 2 and Landsat 8 sensors of invasive plants such as the Nile rose or water hyacinth (
Eichhornia crassipes) with superior sensing properties have presented untapped prospective options
[80]. The spectral reflectance of the Landsat 8 operational land imager OLI was used to distinguish the water hyacinth’s spectral signature from other plants. These indices revealed the highest reflection of the water hyacinth plant compared to other plants.
As each vegetation species has similar spectral signatures, spectral classification of vegetation types in complex environments is difficult
[81]. However, other researchers discovered that water hyacinth has a specific spectral or textural signal that allows it to be distinguished from surrounding native flora
[82]. Textural and phonological variations were efficient approaches for identifying water hyacinth. Another contribution showed that hyperspectral data constitute an appropriate strategy for detecting invasive plants based on differences in spectral signatures
[83]. Water hyacinth displayed higher NIR reflectance values than related plant species and water, owing to the high reflectance of the internal spongy leaf structure (700–1000 nm)
[84]. The spectral signatures of hyacinth detected typical characteristics with low reflectance in the visible part of the spectrum due to high concentrations of chlorophyll-a, which is an indicator of healthy aquatic vegetation conditions
[85].