Spinach leaf disease identification based on deep learning techniques

Spinach is a high-nutritional-value vegetable. However, global warming, climate change, and other essential elements, such as pests, have a negative impact on spinach growth and produce many diseases that limit and destroy the production of healthy crops, making early and correct identification of these diseases critical. Many studies in recent years have employed deep learning models to automatically diagnose vegetable leaf diseases. However, many of these methods are based on separate deep learning architectures, ignoring the different effects of different channels and spatial location relationships in the feature map on classification, resulting in insufficient image representation. Therefore, this paper proposes an integrated deep learning system for automatic recognition of spinach leaf disease. First, convolutional neural networks (CNNs) are used to repeatedly train and extract important features from shallow layers to enhance the recognition of the network. Second, a novel spatial attention mechanism is introduced, which decomposes the two-dimensional space into horizontal and vertical dimensions. While paying attention to the local area, the attention weight can be mapped to the adjacent position of the lesion area, providing rich features for the model. Finally, the outer product matrix of the attention mechanism is kerneled to model the nonlinear relationship between channels using a Sigmoid kernel function, and the final classification is completed by a softmax layer. Experimental results demonstrate that the correct classification rate is 95.12 on our test set. This finding demonstrates the reliability of the proposed hybrid model as an automatic identifier of spinach plant diseases, which can significantly contribute to providing better solutions for identifying other crop diseases in the agricultural field.