SoyNet: Soybean leaf diseases classification

Abstract According to studies, the human population would cross 9 billion by 2050 and the food demand would increase by 60%. Therefore, increasing and improving the quality of the crop yield is a major field of interest. Recently, infectious biotic and abiotic diseases reduce the potential yield by an average of 40% with many farmers in the developing world experiencing yield losses as high as 100%. Farmers worldwide deal with the issue of plant diseases diagnosis and their proper treatment. With advancements of technology in precision agriculture, there has been quite a few works done for plant diseases classification although, the performances of the existing approaches are not satisfactory. Moreover, most of the previous works fail to accurately segment leaf part from the whole image especially when an image has complex background. Thus, a computer vision approach is proposed in order to address these challenges. The proposed approach consists of two modules. The first module extracts leaf part from whole image by subtracting complex background. The second module introduces a deep learning convolution neural network (CNN), SoyNet, for soybean plant diseases recognition using segmented leaf images. All the experiments are done on “Image Database of Plant Disease Symptoms” having 16 categories. The proposed model achieves identification accuracy of 98.14% with good precision, recall and f1-score. The proposed method is also compared with three hand-crafted features based state-of-the-art methods and six popularly used deep learning CNN models namely, VGG19, GoogleLeNet, Dense121, XceptionNet, LeNet, and ResNet50. The obtained results depict that the proposed method outperforms nine state-of-the-art methods/models.