ENHANCING POTATO LEAF DISEASE DETECTION USING HYBRID CNN ARCHITECTURES: ACOMPARATIVE STUDY OF VGG16, RESNET50, AND MOBILENETV2

Abstract

An investigation studies deep learning methods that use image classification to automate detecting potato leaf diseases. The research combines a created Convolutional Neural Network along with pre-trained VGG16 MobileNetV2 and ResNet50 models to increase performance level on a restricted agricultural dataset through transfer learning approaches for early disease identification in improving crop sustainability and yield.

The research methodology includes data preparation before training models to evaluate their performance by standardized measurement of accuracy alongside precision, recall, F1-score and AUC. The pre-trained ResNet50 achieves the highest accuracy and best robustness along with pre-trained models surpassing the custom CNN based on results produced by performance tests. MobileNetV2 demonstrates the best results for achieving computational efficiency at optimal performance levels. The research has shown that deep learning technology offers real-time smartphone-based agricultural disease diagnosis capabilities to farmers through its viable solutions. The research proposes innovative next steps which include augmenting the dataset to improve its diversity as well as enhancing lightweight model performance and mobile application deployment for practical agricultural use.