L-CSMS: novel lightweight network for plant disease severity recognition

The automatic identification and diagnosis of plant disease severity are still challenging in the agricultural information system. In this study, a novel lightweight convolutional neural network (CNN)-based network with channel shuffle operation and multiple-size module (L-CSMS) is proposed for plant disease severity recognition. Specifically, the proposed stacked block consists of residual learning, channel shuffle operation, and multiple-size convolutional modules. The main contributions of this paper include the following: a lightweight and accurate network for practical plant disease severity diagnosis system is designed; it is the first attempt to incorporate the channel shuffle operation and the multiple-size convolution module into the building block as a stacked topology. Finally, the proposed lightweight CNNs-based model achieves a competitive performance over the previous works (such as ShuffleNet, MobileNet) with the accuracy of 0.906 and 0.979 on the plant disease severity dataset and PlantVillage dataset, respectively. Additionally, extensive experiments are conducted to demonstrate that the proposed method is effective for plant disease diagnosis.