Automatic Detection of Tomato Diseases Using Deep Transfer Learning

Abstract

Global food production is being strained by extreme weather conditions, fluctuating temperatures, and geopolitics. Tomato is a staple agricultural product with tens of millions of tons produced every year worldwide. Thus, preserving the tomato plant from diseases will go a long way in reducing economical loss and boost output. Technological innovations have great potential in facilitating disease detection and control. More specifically, artificial intelligence algorithms in the form of deep learning methods have established themselves in many real-life applications in a wide range of disciplines (e.g., medicine, agriculture, or facial recognition, etc.). In this paper, we aim at applying deep transfer learning in the classification of nine tomato diseases (i.e., bacterial spot, early blight, late blight, leaf mold, mosaic virus, septoria leaf spot, spider mites, target spot, and yellow leaf curl virus) in addition to the healthy state. The approach in this work uses leaf images as input, which is fed to convolutional neural network models. No preprocessing, feature extraction, or image processing is required. Moreover, the models are based on transfer learning of well-established deep learning networks. The performance was extensively evaluated using multiple strategies for data split and a number of metrics. In addition, the experiments were repeated 10 times to account for randomness. The ten categories were classified with mean values of 99.3% precision, 99.2% F1 score, 99.1% recall, and 99.4% accuracy. Such results show that it is highly feasible to develop smartphone-based applications that can aid plant pathologists and farmers to quickly and accurately perform disease detection and subsequent control.

JujubeNet: A high-precision lightweight jujube surface defect classification network with an attention mechanism

“…In DL methods, Accuracy, Recall, Precision, and F1-scores are significant metrics for evaluating the merits of classification models ( Khasawneh et al., 2022 ). The accuracy rate is the percentage of the correct samples predicted by the model to the total number of samples; the recall rate is the percentage of the model correctly predicted as a positive sample to the total number of positive samples; the precision rate is the percentage of the number of positive samples predicted by the model that genuinely belongs to positive samples, and the F1-score is the best balance point that the model measures both the precision rate and the recall rate and achieves, and this value also reflects the overall performance of the model more comprehensively.…”

Monitoring Tomato Leaf Disease through Convolutional Neural Networks

“…Some authors have developed real-time models to accelerate the process of disease detection in plants [43,44]. Other authors have created models that contribute to the early detection of plant diseases [45,46]. In [47], the authors make use of images of tomato leaves to discover different types of diseases.…”

Agroclimatic and Phytosanitary Events and Emerging Technologies for Their Identification in Avocado Crops: A Systematic Literature Review

“…Advances in the adaptation of ML and DL model-based techniques [89], along with the development of IoT prototypes for crop data capture [21], are the two fronts that should continue to be studied to obtain improvements in avocado production. Among the applications of emerging technologies in agricultural applications, there is a high use of models based on CNN classifiers for the detection of diseases in the leaves of trees, with research working on the preparation of these models through transfer learning to detect diseases in products such as corn, apples, and tomatoes, with accuracy measurements between 98.6% and 99.4% obtained in the testing of classifiers [90][91][92]. These advances are relevant for the use of emerging technologies in the identification of APEs, and in the specific case of avocado crops, the use of data from the crop and the use of pretrained models with related data should be prioritized since there is a bias in the application of transfer learning to pretrained models for other agricultural products.…”