Detecting and Pinpointing the Diseased Areas of Mango Leaves using RNN and SegNet

The agricultural sector, particularly mango production, is a considerable challenge due to variety leaf diseases that can adversely affect yields and quality. This article presents a novel approach using Recurrent Neural Networks (RNNs) and SegNet to identify and localization of diseased areas of mango leaves. It uses dataset of 4069 images divided into eight diseases: Anthracnose, Gall Midge, Sooty Mould, Cutting weevil, Die Back, Powdery Mildew, and Bacterial canker and Healthy. This model includes VGG19 for feature extraction, RNN for sequencing analysis, and SegNet for image segmentation. The Python-based system includes an intuitive Graphical User Interface (GUI) that identifies the illness type, assesses the degree of affection, and recommends appropriate treatments. The model's performance was evaluated using metrics like accuracy, loss, sensitivity, ROC curves, and confusion matrices. The results show a high level of accuracy and robustness in the detection and localization of mango leaf diseases with general accuracy and different sensitivity values for all disease categories. User-Friendly GUI ensures it is easy to use. Solutions are effective and inexpensive. Future work will focus on including model improvements, actual monitoring and integration into mobile devices, improving accessibility and user friendliness.