SwinCNN: A Hybrid Deep Learning Architecture for Accurate Cotton Disease Prediction

Cotton crop diseases can have a substantial impact on crop productivity and quality in the agricultural industry. The accurate and timely identification of these disorders is critical for optimal disease management. In this work, we present SwinCNN which is a new hybrid Deep Learning (DL) architecture for reliable cotton disease prediction. The SwinCNN architecture combines the strengths of Convolutional Neural Networks (CNN) and the Swin Transformer (ST), a recent advancement in self-attention (SA) mechanisms for computer vision tasks. We use a publicly available dataset of cotton crop images with various diseases to train and evaluate our proposed architecture. Our results show that the SwinCNN architecture outperforms both CNN and ST models individually, achieving an overall accuracy of 99.65% on the test set. The proposed architecture can be used as a valuable tool in smart farming to detect and predict cotton crop diseases, enabling farmers to take necessary actions to mitigate crop losses.