TOMATO LEAF MOBILE AN ANDROID BASED REAL TIME DISEASE DETECTION SYSTEM

Abstract

Tomato leaf diseases are very critical to the agricultural production, quality of the crop and global food security. Thus, prompt and precise disease identification is crucial for sustainable farming methods. Manual inspection is labour-intensive, subjective and often ineffective and is the major drawback for large-scale agriculture, where intelligent automated diagnosis is needed. The proposed approach is based on the publicly available PlantVillage tomato leaf disease 1 dataset, which consists of labeled images of healthy and diseased tomato leaves collected under controlled conditions, spanning various categories of diseases. The image scaling, normalisation, data augmentation, one-hot encoding and train-validation-test partitioning are part of the processing pipeline to enhance the representation of the features, model generalisation and reduce the over-fitting effect. Implementations of different DL architectures such as Custom CNN, ResNet50, DenseNet121, MobileNetV2, Xception, NASNetLarge, and Hybrid Ensemble of Xception and NASNetLarge are done for benchmarking. The modified MobileNetV3, incorporating Inverted Residual Blocks and Efficient Channel Attention mechanisms, on the other hand, improves the efficiency of lightweight feature extraction. The accuracy, precision, recall, F1 score, confusion matrix, ROC analysis and learning curves of training-validation are used to assess the performance. The experimental results revealed that the best performance in terms of accuracy, precision, recall and F1-score was obtained by the Hybrid Ensemble with 99.9%, 99.9%, 99.9% and 99.9% respectively. The interpretability and usefulness of the visualisation was enhanced by the GradCAM visualisation, while the diagnosis interface and treatment suggestions along with the GTTS voice assistance were rendered more useful and practical through Flask. Finally, the proposed framework provides the intelligent agriculture applications with high accuracy, efficiency and interpretability of tomato leaf diseases detection.