Plant Diseases and Pests are a major challenge in the agriculture sector. An accurate and a faster detection of diseases and pests in plants could help to develop an early treatment technique while substantially reducing economic losses. Recent developments in Deep Neural Networks have allowed researchers to drastically improve the accuracy of object detection and recognition systems. In this paper, we present a deep-learning-based approach to detect diseases and pests in tomato plants using images captured in-place by camera devices with various resolutions. Our goal is to find the more suitable deep-learning architecture for our task. Therefore, we consider three main families of detectors: Faster Region-based Convolutional Neural Network (Faster R-CNN), Region-based Fully Convolutional Network (R-FCN), and Single Shot Multibox Detector (SSD), which for the purpose of this work are called "deep learning meta-architectures". We combine each of these meta-architectures with "deep feature extractors" such as VGG net and Residual Network (ResNet). We demonstrate the performance of deep meta-architectures and feature extractors, and additionally propose a method for local and global class annotation and data augmentation to increase the accuracy and reduce the number of false positives during training. We train and test our systems end-to-end on our large Tomato Diseases and Pests Dataset, which contains challenging images with diseases and pests, including several inter- and extra-class variations, such as infection status and location in the plant. Experimental results show that our proposed system can effectively recognize nine different types of diseases and pests, with the ability to deal with complex scenarios from a plant's surrounding area.

A Robust Deep-Learning-Based Detector for Real-Time Tomato Plant Diseases and Pests Recognition

With increasing demand to support and accelerate progress in breeding for novel traits, the plant research community faces the need to accurately measure increasingly large numbers of plants and plant parameters. The goal is to provide quantitative analyses of plant structure and function relevant for traits that help plants better adapt to low-input agriculture and resource-limited environments. We provide an overview of the inherently multidisciplinary research in plant phenotyping, focusing on traits that will assist in selecting genotypes with increased resource use efficiency. We highlight opportunities and challenges for integrating noninvasive or minimally invasive technologies into screening protocols to characterize plant responses to environmental challenges for both controlled and field experimentation. Although technology evolves rapidly, parallel efforts are still required because large-scale phenotyping demands accurate reporting of at least a minimum set of information concerning experiment...

https://horticulture.wisc.edu/wp-content/uploads/sites/20/2013/09/Fiorani-and-Schurr-2013-Future-Scenarios-of-Plant-Phenotyping2.pdf

Future scenarios for plant phenotyping.

Traditional imagery—provided, for example, by RGB and/or NIR sensors—has proven to be useful in many agroforestry applications. However, it lacks the spectral range and precision to profile materials and organisms that only hyperspectral sensors can provide. This kind of high-resolution spectroscopy was firstly used in satellites and later in manned aircraft, which are significantly expensive platforms and extremely restrictive due to availability limitations and/or complex logistics. More recently, UAS have emerged as a very popular and cost-effective remote sensing technology, composed of aerial platforms capable of carrying small-sized and lightweight sensors. Meanwhile, hyperspectral technology developments have been consistently resulting in smaller and lighter sensors that can currently be integrated in UAS for either scientific or commercial purposes. The hyperspectral sensors’ ability for measuring hundreds of bands raises complexity when considering the sheer quantity of acquired data, whose usefulness depends on both calibration and corrective tasks occurring in pre- and post-flight stages. Further steps regarding hyperspectral data processing must be performed towards the retrieval of relevant information, which provides the true benefits for assertive interventions in agricultural crops and forested areas. Considering the aforementioned topics and the goal of providing a global view focused on hyperspectral-based remote sensing supported by UAV platforms, a survey including hyperspectral sensors, inherent data processing and applications focusing both on agriculture and forestry—wherein the combination of UAV and hyperspectral sensors plays a center role—is presented in this paper. Firstly, the advantages of hyperspectral data over RGB imagery and multispectral data are highlighted. Then, hyperspectral acquisition devices are addressed, including sensor types, acquisition modes and UAV-compatible sensors that can be used for both research and commercial purposes. Pre-flight operations and post-flight pre-processing are pointed out as necessary to ensure the usefulness of hyperspectral data for further processing towards the retrieval of conclusive information. With the goal of simplifying hyperspectral data processing—by isolating the common user from the processes’ mathematical complexity—several available toolboxes that allow a direct access to level-one hyperspectral data are presented. Moreover, research works focusing the symbiosis between UAV-hyperspectral for agriculture and forestry applications are reviewed, just before the paper’s conclusions.

/pdf/hyperspectral-imaging-a-review-on-uav-based-sensors-data-h7zu88r0tz.pdf

Hyperspectral Imaging: A Review on UAV-Based Sensors, Data Processing and Applications for Agriculture and Forestry

Early and accurate detection and diagnosis of plant diseases are key factors in plant production and the reduction of both qualitative and quantitative losses in crop yield. Optical techniques, such as RGB imaging, multi- and hyperspectral sensors, thermography, or chlorophyll fluorescence, have proven their potential in automated, objective, and reproducible detection systems for the identification and quantification of plant diseases at early time points in epidemics. Recently, 3D scanning has also been added as an optical analysis that supplies additional information on crop plant vitality. Different platforms from proximal to remote sensing are available for multiscale monitoring of single crop organs or entire fields. Accurate and reliable detection of diseases is facilitated by highly sophisticated and innovative methods of data analysis that lead to new insights derived from sensor data for complex plant-pathogen systems. Nondestructive, sensor-based methods support and expand upon visual and/or molecular approaches to plant disease assessment. The most relevant areas of application of sensor-based analyses are precision agriculture and plant phenotyping.

https://apsjournals.apsnet.org/doi/pdf/10.1094/PDIS-03-15-0340-FE

Plant Disease Detection by Imaging Sensors - Parallels and Specific Demands for Precision Agriculture and Plant Phenotyping.

Reliable, precise and accurate estimates of disease severity are important for predicting yield loss, monitoring and forecasting epidemics, for assessing crop germplasm for disease resistance, and for understanding fundamental biological processes including co-evolution. Disease assessments that are inaccurate and/or imprecise might lead to faulty conclusions being drawn from the data, which in turn can lead to incorrect actions being taken in disease management decisions. Plant disease can be quantified in several different ways. This review considers plant disease severity assessment at the scale of individual plant parts or plants, and describes our current understanding of the sources and causes of assessment error, a better understanding of which is required before improvements can be targeted. The review also considers how these can be identified using various statistical tools. Indeed, great strides have been made in the last thirty years in identifying the sources of assessment error inherent to visual rating, and this review highlights ways that assessment errors can be reduced—particularly by training raters or using assessment aids. Lesion number in relation to area infected is known to influence accuracy and precision of visual estimates—the greater the number of lesions for a given area infected results in more overestimation. Furthermore, there is a widespread tendency to overestimate disease severity at low severities (<10%). Both interrater and intrarater reliability can be variable, particularly if training or rating aids are not used. During the last eighty years acceptable accuracy and precision of visual disease assessments have often been achieved using disease scales, particularly because of the time they allegedly save, and the ease with which they can be learned, but recent work suggests there can be some disadvantages to their use. This review considers new technologies that offer opportunity to assess disease with greater objectivity (reliability, precision, and accuracy). One of these, visible light photography and digital image analysis has been increasingly used over the last thirty years, as software has become more sophisticated and user-friendly. Indeed, some studies have produced very accurate estimates of disease using image analysis. In contrast, hyperspectral imagery is relatively recent and has not been widely applied in plant pathology. Nonetheless, it offers interesting and potentially discerning opportunities to assess disease. As plant disease assessment becomes better understood, it is against the backdrop of concepts of reliability, precision and accuracy (and agreement) in plant pathology and measurement science. This review briefly describes these concepts in relation to plant disease assessment. Various advantages and disadvantages of the different approaches to disease assessment are described. For each assessment method some future research priorities are identified that would be of value in better understanding the theory of disease assessment, as it applies to improving and fully realizing the potential of image analysis and hyperspectral imagery.

Plant Disease Severity Estimated Visually, by Digital Photography and Image Analysis, and by Hyperspectral Imaging

Citrus canker is caused by the bacterial pathogen Xanthomonas axonopodis pv. citri and infects several citrus species in wet tropical and subtropical citrus growing regions. Accurate, precise, and reproducible disease assessment is needed for monitoring epidemics and disease response in breeding material. The objective of this study was to assess reproducibility of image analysis (IA) for measuring severity of canker symptoms and to compare this to visual assessments made by three visual raters (VR1-3) for various symptom types (lesion numbers, % area necrotic, and % area necrotic+chlorotic), and to assess inter- and intra-VR reproducibility. Digital images of 210 citrus leaves with a range of symptom severity were assessed on two separate occasions. IA was more precise than VRs for all symptom types (inter-assessment correlation coefficients, r, for lesion numbers by IA = 0.99, by VRs = 0.89 to 0.94; for %, r for % area necrotic+chlorotic for IA = 0.97 and for VRs = 0.86 to 0.89; and r for % area necrotic for IA = 0.96 and for VRs = 0.74 to 0.85). Accuracy based on Lin's concordance coefficient also followed a similar pattern, with IA being most consistently accurate for all symptom types (bias correction factor, Cb = 0.99 to 1.00) compared to visual raters (Cb = 0.85 to 1.00). Lesion number was the most reproducible symptom assessment (Lin's concordance correlation coefficient, ρc, = 0.76 to 0.99), followed by % area necrotic+chlorotic (ρc = 0.85 to 0.97), and finally % area necrotic (ρc = 0.72 to 0.96). Based on the "true" value provided by IA, precision among VRs was reasonable for number of lesions per leaf (r = 0.88 to 0.94), slightly less precision for % area necrotic+chlorotic (r = 0.87 to 0.92), and poorest precision for % area necrotic (r = 0.77 to 0.83). Loss in accuracy was less, but showed a similar trend with counts of lesion numbers (Cb = 0.93 to 0.99) which was more consistently accurately reproduced by VRs than either % area necrotic (Cb = 0.85 to 0.99) or % area necrotic+chlorotic (Cb = 0.91 to 1.00). Thus, visual raters suffered losses in both precision and accuracy, with loss in precision estimating % area necrotic being the greatest. Indeed, only for % area necrotic was there a significant effect of rater (a two-way random effects model ANOVA returned a P 20% area (e.g., 25, 30, 35, 40, etc.), yet VRs were prepared to estimate disease of <1% area, and at 1% increments up to 20%. There was a linear relationship between actual disease (IA assessments) and VRs. IA appears to provide a highly reproducible way to assess canker-infected leaves for disease, but symptom characters (symptom heterogeneity, coalescence of lesions) could lead to discrepancies in results, and full automation of the system remains to be tested.

https://apsjournals.apsnet.org/doi/pdf/10.1094/PDIS-92-4-0530

Visual Rating and the Use of Image Analysis for Assessing Different Symptoms of Citrus Canker on Grapefruit Leaves

Dynamics of dispersal of the bacteria that causes citrus canker (Xanthomonas axonopodis pv. citri) were assessed in simulated wind-driven rain splash. The wind/rain-splash events were simulated using electric blowers to generate turbulent wind (15 to 20 m s-1) and sprayer nozzles to produce water droplets entrained in the wind flow. The splash was blown at an inoculum source of canker-infected trees 1 m downwind. The splash downwind of the source of the infected trees was collected by vertical panel samplers and funnel samplers. The duration over which bacteria were dispersed in spray was assessed in continuous wind at intervals from 0 to 52 h after commencing the simulated rain splash event. In one experiment on 11 February 2003, a total of 1.48 × 106 bacteria were collected by panels 1 m downwind from the inoculum source during the first 10 min of dispersal, but the numbers declined to 3.60 × 105 bacteria after 1 h and ranged between 1.42 × 105 and 1.93 × 104 up to 52 h. In a more detailed stud...

https://apsjournals.apsnet.org/doi/pdf/10.1094/PD-89-0071

Effect of Simulated Wind-Driven Rain on Duration and Distance of Dispersal of Xanthomonas axonopodis pv. citri from Canker-Infected Citrus Trees.

Distribution of Biological Control Agents of Leafy Spurge (Euphorbia esulaL.) in the United States: 1988–1996

Citrus canker (caused by Xanthomonas citri subsp. citri) is a destructive disease, reducing yield and rendering fruit unfit for fresh sale. Accurate assessment of citrus canker severity and other diseases is needed for several purposes, including monitoring epidemics and evaluation of germplasm. We compared measurements of citrus canker severity (percent area infected) from automated image analysis to visual estimates by raters and true values using images from five leaf samples (65, 123, 50, 50, and 200 leaves; disease severity from 0 to 60%). Severity on leaves was measured by automated image analysis by (i) basing threshold values on a presample of leaves, or (ii) replacing healthy leaf color on a leaf-by-leaf basis before automating image analysis. Samples 1 to 4 were assessed by three trained plant pathologists, and sample 5 was assessed by an additional 25 raters. Healthy leaf area color replacement gave the most consistent agreement with the true severity data. Using color replacement, agreement with true values based on Lin's concordance correlation coefficient (ρc) was 0.93, 0.79, 0.71, 0.85, and 0.89 for each of the samples, respectively. The range and consistency of agreement was generally less good for automated thresholds based on a presample (ρc = 0.35-0.90) or visual raters (ρc = 0.30-0.94). The constituents of agreement (precision and accuracy) showed similar trends. No one rater or method was best for every leaf sample, but replacing healthy color in each leaf with a standard color before automation of image analysis improved agreement, and was relatively quick (20 s per image). The accuracy and precision of automated image analysis of citrus canker severity can be comparable to unaided, direct visual estimation by many raters.

/pdf/automated-image-analysis-of-the-severity-of-foliar-citrus-46enqyiho2.pdf

P. E. Parker

Papers

Plant Disease Severity Estimated Visually, by Digital Photography and Image Analysis, and by Hyperspectral Imaging

Visual Rating and the Use of Image Analysis for Assessing Different Symptoms of Citrus Canker on Grapefruit Leaves

Effect of Simulated Wind-Driven Rain on Duration and Distance of Dispersal of Xanthomonas axonopodis pv. citri from Canker-Infected Citrus Trees.

Distribution of Biological Control Agents of Leafy Spurge (Euphorbia esulaL.) in the United States: 1988–1996

Automated Image Analysis of the Severity of Foliar Citrus Canker Symptoms