Detection of soybean pests and diseases through machine learning techniques

Abstract

Soybean, a vital source of protein and vegetable oil, plays a significant role in the economic growth of producing countries. However, diseases and pest infestation pose a substantial threat to soybean yield. Early detection of these issues on soybean leaves is crucial for preventing or reducing production losses. Machine learning and deep learning methods have shown promise in detecting soybean diseases. In this study, we investigated commonly used models for plant image classification, focusing on soybean disease and pest identification. Six models were selected, including three simple machinelearning models and three deep-learning models. To evaluate their performance, we used classification accuracy, precision, recall, and F-measure metrics. Our results surpassed those of previous studies, achieving improved accuracy in detecting soybean diseases and pests. Notably, the disease data set outperformed the pest data set, despite the latter being larger. Among the algorithms tested, VGG-16, a deep learning algorithm, demonstrated superior performance. The following are the classification accuracies achieved for the pest and disease data sets, respectively, using different algorithms: Support Vector Machine (88% and 92%), Random Forest (83% and 95%), K Nearest Neighbors (76% and 74%), VGG-16 (95% and 99%), VGG-19 (94% and 98%), and our custom CNN algorithm, ViewNet (89.5% and 75%). By leveraging 10-fold cross-validation, a widely used technique in machine learning, we ensured reliable and robust evaluations of the models. These findings contribute to the advancement of agricultural practices by providing insights into effective machine learning and deep learning approaches for soybean disease and pest detection.