A Comprehensive Comparison of MobileNet, ResNet50 and InceptionV3 for Efficient Plant Pathology Detection

Plant diseases have a strong impact on agricultural productivity due to economic factors and a reduction in crop quality. This work focuses on the classification of apple leaves into four classes: healthy, rust-infected, scab-infected, and infected by both diseases, using the Plant Pathology FGVC7-2020 dataset that contains 3,642 images in total. The work involves the analysis of three sophisticated deep learning architectures: ResNet50, MobileNet and InceptionV3.It turned out that MobileN et showed the highest performance, reaching a 92% accuracy rate; it was followed by ResNet50 with 75% accuracy and InceptionV3 at 73%, hence underlining its better generalization capability and efficiency in classifying. We discuss the proposed methodology, which includes data preprocessing techniques, experimental results and final conclusions, is discussed in detail. These results underline the fundamental importance of determining an appropriate neural network architecture for the recognition of plant diseases, which is of prime importance to improve agricultural productivity.