PlaNet: a robust deep convolutional neural network model for plant leaves disease recognition

Researchers are looking for new ideas that can greatly increase the amount of food grown while also cutting costs. For precision agriculture to work, pests, weeds, and diseases must be easy to find and identify on plant leaves. Most plant diseases have symptoms that can be seen, and plant pathologists now agree that infected plant leaves are the best way to find them. This is a good use for computer-aided diagnostic systems because diagnosing diseases manually (hands and eyes) takes a long time, and the effectiveness of diagnostic treatment depends on how well the pathologist does (intra-observer variability). Based on what we need right now, we need a model that doesn't need much pre-processing and doesn't need manual functional (feature) extraction. So, in this study, methods for identifying and classifying plant diseases from leaf images taken at different resolutions are described that are based on deep learning. Using Deep Convolutional Neural Network (DCNN) image analysis, the main goal of this research is to help tell the difference between healthy and unhealthy leaves. We have also come up with a new model called PlaNet. Its performance has been compared to that of other common CNN models. We did a lot of testing and verification on 18 well-known CNN models based on deep learning, including one ensemble model made up of the five best models. We did this on four different combinations of three well-known standard benchmark datasets. The suggested PlaNet model has been tested, and the results show that it works in a highly efficient manner. It has been found that, among all the testing experiments, the best average-wise performance achieved was up to 97.95% accuracy, 0.9752 AUC, 0.9686 F1-score, 0.9707 sensitivity, 0.9576 precision, and 0.9456 specificity, respectively. So, the results of the tests show that the proposed system, which is based on deep learning, can classify different types of plant leaves quickly and accurately. When used alone or with other datasets (four different combinations of three datasets), the suggested model works very well. This shows that the proposed model is more flexible, general, and scalable than other approaches that are already in use. The fact that it must be able to accurately process a single image of a plant leaf in less than a second demonstrates its real-time capabilities.