LWCNN: a lightweight convolutional neural network for agricultural crop protection

Abstract

Automatic identification of plant diseases is critical for agricultural crop protection so as to enhance the crop yield. The recent advances in deep learning and image processing gives hope for the development of efficient algorithms to address this issue. In this manuscript, we make use of these schemes to develop a Light-Weight Convolutional Neural Network (LWCNN) for identifying diseases in the leaves and ears of pearl millets. Although many models exist in the literature, the total number of parameters employed by our model is far fewer, by an order of thousand as compared to many other light-weight networks such as MobileNet(v2), EfficientNet, NASNet etc. Hence our scheme can be employed and run directly on devices with much lesser compute power. It is noteworthy that despite using few parameters, the proposed model achieves an accuracy of 97.4% in detecting the existence of the downy mildew disease in pearl millets, and takes the least time for both training and testing as compared to other models. To eliminate most of the pre-processing steps and to make our system suitable for on-field detection, we explore three single stage object detectors namely SSD, YOLOv3 and RetinaNet which localize and classify multiple instances of healthy and diseased leaves and ears in the image. We present a comparative analysis of the models and our experiments indicate that SSD is most suitable outperforming the other two models by a significant margin.