The detection of flaws in food crops, such as potatoes, may benefit greatly from the use of machine vision and image processing techniques. There has been an uptick in the use of image processing and AI in agriculture, particularly for the purposes of detecting and categorising plant and fruit pests and illnesses. Manual interpretation of these leaf diseases may be time-consuming and labor-intensive, but they have a significant impact on potato quality and yield due to problems like early blight and late blight. Effective and automated diagnosis of these diseases during the budding period may help improve potato crop output since it demands a very high degree of competence. Several models have been presented in the past for early detection of plant diseases. The goal of this study is to demonstrate TuberNet, a deep learning (DL) method to illness recognition in potatoes. In particular, the research employs the TuberNet in order to recognise different potato leaf diseases, hence introducing an end-to-end training-oriented strategy. To improve the approach's identification capabilities and better identify many infections, we adopt a spatial-channel attention strategy to zero in on the damaged regions. In order to deal with class-imbalanced samples and boost the network's generalisation capacity, dense layers are added to the end of the model structure to increase the feature selection power of the model, and transfer learning is used. The model is put through its paces on the open and difficult Plant Village dataset, which contains photographs captured under a wide variety of lighting and background circumstances, as well as those showing a wide range of leaf coloration. Experiments on the model's accuracy for disease classification in potato plant leaves and its ability to handle distorted data robustly have been proven. As a result, the suggested model tool may increase farmers' profits and crop yields..

Deep Learning; TuberNet; Plant Disease Diagnosis; Transfer Learning; Potato Plant; Agriculture.