CST: Convolutional Swin Transformer for detecting the degree and types of plant diseases

In modern agriculture, detecting and identifying plant diseases is a severe difficulty. Because plant diseases can cause major economic damage and endanger food safety. Due to the advancement of artificial intelligence technologies, numerous research projects have utilized photos in various methods to detect plant illnesses. However, noise and other factors are unavoidable in image capture, affecting detection accuracy. To overcome this problem, we presented a Convolutional Swin Transformer (CST) based on the Swin Transformer to recognize the degree and kind of disease. Using a novel convolutional design, the CST model proposed in this research can achieve both high detection accuracy and outstanding robustness. CST has an accuracy of 0.909 and 0.922 when detecting plant disease in a natural environment, 0.975 when identifying disease in a controlled environment, and 0.982 when identifying disease categories. It maintains accuracy of 0.795 even when detecting photos with 30% salt noise. The research results are likely to be used in natural plant disease monitoring systems to make them more effective and reliable. • A class of networks is designed to detect the degree and type of plant diseases. • Utilizing Swin Transformer as the network’s backbone. • Validating superior performance across three datasets. • The maximum improvement in accuracy is 25.2%. • Excellent robustness in detecting noisy images.