Diagnosis of contrast improved common maize leaf diseases using U2Net based deep learning model

India is the 14 th largest exporter of maize worldwide thanks to its strategic and geographical location with round the year production. However, the crop is affected by more than 65 types of disease as is reported from different regions of India. This has the catastrophic impact on not just the food security but also economy of the country as whole. Timely and low-cost system for the early diagnosis of crop disease at the hands of cultivators becomes the game charger in the crop management system. With the availability of smart phone at the hands of farmers, deep learning has the important role in the field of agriculture. In our study we have explored the DCNN as the crop disease classification and detection system pertaining to the corn leaf by the application of U 2 Net and contrast improved leaf images as an image processing technique. At the initial stage of study, we examined different state of the art CNN for original field images of three different disease in CD&S dataset viz. GLS, NLB and NLS for various performance indicators. MobileNet is then selected for further investigation as the model of choice due to its light weight and quick response time beside accuracy and loss. We then perform U 2 Net based background removal approach to free images form the clutter, noise etc. before making three different contrast enhancement transformation. MobileNet CNN is then trained and tested over four different datasets prepared after data augmentation and other preprocessing techniques. We found that Adaptive Multiscale Retinex (AMSR) gives the best accuracy of 98.08 and 98.98 on validation and test dataset. Overall, the study explored the removal of background clutter from the sample diseased leaf image using U 2 Net and then enhancing the vividness of diseased liaison with contrast improvement techniques thus helping the DCCN to focus better on the diseased part which in turn can improve the efficiency in terms of speed and accuracy of detection and classification of crop diseases.