Deep Learning Based Leaf Detection and Disease Prediction System using ResNet-50 and OpenCV for Precision Agriculture
Abstract
The focus of this research paper is the problem of proper leaf identification and disease detection of plants in agricultural practice, where manual inspections are time-consuming, subjective and cannot be scaled. It suggests an advanced system (Deep Learning-based Leaf Detection and Disease Prediction System) based on resnet-50 with the transfer learning method and preprocesses the image with OpenCV (to increase the accuracy of feature extraction and classification). Noise removal, edge detection and contour analysis are preprocessing methods to enhance the separation of leaf regions, whereas data augmentation can enhance the generalization of models. The model is trained using the Cassava and Pepper leaf data, with healthy and diseased classes to perform powerful multi-class classification. The results of the experiment are high performance with accuracy, precision, recall and F1-score greater than traditional methods, reaching an accuracy of 99.87%. The system facilitates early detection of disease, crop management and precision agriculture of deployments. This method combines the use of deep neural networks and classical image processing to augment agricultural decision support systems.