Plant leaf disease detection and classification using a lightweight DPCNet model

Plant disease is one of the major causes for an overall decrease in yield and reduces the profit from the farming segment. Convolutional Neural Networks (CNN) has the potential to classify multiple classes with automatic feature extraction capability and can be light weight to run on edge devices. Therefore, this research work concerns with development of novel lightweight CNN architectures for leaf disease classification. To acquire significant features in lesser number of trainable parameters, a novel microarchitecture module with Depthwise convolution layer, Pointwise Convolution layer and Conventional Convolution layer is proposed. These modules are arranged linearly in three branches to obtain colour, shape and texture features and the developed CNN architecture is called "DPCNet model". Further, to extract texture information, a novel hybrid CNN architecture called DPCNet is proposed by combining features from Gray Level Co-occurrence Matrix (GLCM) with CNN feature maps. The proposed architecture is analyzed on plant village dataset. Among the various CNN models, proposed DPCNet delivers highest validation accuracy and least validation loss of 95.78% and 0.1478 respectively with 1,27,425 trainable parameters than the other existing CNN models.