Semantic Segmentation and Transfer Learning based Disease Classification in Black Gram Plants

Abstract

Smart farming, also known as precision agriculture, entails the integration of computing technologies into agriculture to promote sustainable and environment friendly practices. The high prevalence of plant diseases significantly affects the crop quality and yield, creating a need for a supporting system. Such systems should identify plant diseases more effectively and provide recommendations for the required amount of fertilizers for the particular disease. Hence, in this article, a framework for leaf segmentation and disease classification is proposed, particularly for black gram plants. This system uses advanced deep learning algorithms to segment the plant leaves and classify them into different disease categories. Real-time images have a complex background, which is similar to the leaf, and may mislead the disease recognition algorithms. To overcome this challenge, a semantic segmentation algorithm, which is DeepLabv3+, is proposed to segment leaf regions from the input images. The ResNet-18 architecture is utilized to serve as the backbone of the DeepLabv3+ layers. And then EfficientNet- B0 model is used for the disease classification. The experimental results showed that this combination for both segmentation and classification tasks achieved an accuracy of 99.72%, a precision of 99.35%, a recall of 99.33% and an F1 score of 99.34%. Finally, for the benefit of the farmers, an application is developed to recognize the diseases and recommend fertilizers for the disease.