TOMMicroNet: Convolutional Neural Networks for Smartphone-Based Microscopic Detection of Tomato Biotic and Abiotic Plant Health Issues.

The image-based detection and classification of plant diseases has become increasingly important to the development of precision agriculture. We consider the case of tomato, a high-value crop supporting the livelihoods of many farmers around the world. Many biotic and abiotic plant health issues impede the efficient production of this crop, and laboratory-based diagnostics are inaccessible in many remote regions. Early detection of these plant health issues is essential for efficient and accurate response, prompting exploration of alternatives for field detection. Considering the availability of low-cost smartphones, artificial intelligence-based classification facilitated by mobile phone imagery can be a practical option. This study introduces a smartphone-attachable 30x microscopic lens, used to produce the novel tomato microimaging dataset of 8500 images representing 34 tomato plant conditions on the upper and lower sides of leaves as well as on the surface of tomato fruits. We introduce TOMMicroNet, a 14-layer convolutional neural network (CNN) trained to classify amongst biotic and abiotic plant health issues, and we compare it against six existing pre-trained CNN models. We compared two separate pipelines of grouping data for training TOMMicroNet, either presenting all data at once or separating into subsets based on the three parts of the plant. Comparing configurations based on cross-validation and F1 scores, we determined that TOMMicroNet attained the highest performance when trained on the complete dataset, with 95% classification accuracy on both training and external datasets. Given TOMMicroNet's capabilities when presented with unfamiliar data, this approach has the potential for the identification of plant health issues.

Non-coding deep learning models for tomato biotic and abiotic stress classification using microscopic images.

Plant disease classification is quite complex and, in most cases, requires trained plant pathologists and sophisticated labs to accurately determine the cause. Our group for the first time used microscopic images (×30) of tomato plant diseases, for which representative plant samples were diagnostically validated to classify disease symptoms using non-coding deep learning platforms (NCDL). The mean F1 scores (SD) of the NCDL platforms were 98.5 (1.6) for Amazon Rekognition Custom Label, 93.9 (2.5) for Clarifai, 91.6 (3.9) for Teachable Machine, 95.0 (1.9) for Google AutoML Vision, and 97.5 (2.7) for Microsoft Azure Custom Vision. The accuracy of the NCDL platform for Amazon Rekognition Custom Label was 99.8% (0.2), for Clarifai 98.7% (0.5), for Teachable Machine 98.3% (0.4), for Google AutoML Vision 98.9% (0.6), and for Apple CreateML 87.3 (4.3). Upon external validation, the model's accuracy of the tested NCDL platforms dropped no more than 7%. The potential future use for these models includes the development of mobile- and web-based applications for the classification of plant diseases and integration with a disease management advisory system. The NCDL models also have the potential to improve the early triage of symptomatic plant samples into classes that may save time in diagnostic lab sample processing.