CoSEV: A cotton disease dataset for detection and classification of severity stages and multiple disease occurrence

In agriculture, the development of early treatment techniques for plant leaf diseases can be significantly enhanced by employing precise and rapid automatic detection methods. Within this realm of research, two common scenarios encountered in real field cases are the identification of different severity stages of diseases and the detection of multiple pathogens simultaneously affecting a single plant leaf. One major challenge faced in this area is the lack of publicly available datasets that contain images captured under these specific conditions. To address this challenge, we present a dataset called CoSEV in this paper. The CoSEV dataset comprises a collection of 496 images of cotton leaves, captured both under controlled conditions and in real-field settings using a smartphone camera. Thge total number of images after applying augmentation techniques is 1151.It covers a diverse range of situations, including multiple stresses co-occurring on a single leaf and the progression of disease severity. The dataset was carefully organized into 5 classes, with 7 categories representing different levels of cotton curl severity and coexisting diseases. To evaluate the effectiveness of the CoSEV dataset, we trained and tested various state-of-the-art detection models. These models were analyzed to assess their performance in accurately identifying and classifying the various diseases and severity stages present in the dataset.