STN-FRCNN: Spatial Transformer Network Augmented Faster R-CNN Network for Crop Disease Classification

The various diseases of crops greatly affect agricultural production. The classification of the particular category of disease of a crop in the early stage helps farmers to choose the proper steps against it. The traditional strategies are not only costly but also time-consuming. Deep-Learning (DL)-based approaches can be applied to tackle these shortcomings. This paper proposes a DL-based crop disease classification model that uses a Faster Region-based Convolutional Neural Network (R-CNN). The Faster R-CNN is selected to perform this disease classification task for its outstanding result to identify objects. A Spatial Transformer Network (STN) is added before the Faster R-CNN network to enhance the invariance of convolution functions in case of rotated and shared images of any database. The proposed model outperforms the majority of the time compared to other examined approaches on three standard datasets, i.e., Plant Village (99.36%), Tomato (99.09%), and Grape (99.96%).