COMPARISON OF DETECTION WITH TRANSFER LEARNING ARCHITECTURE RESTNET18, RESTNET50, RESTNET101 ON CORN LEAF DISEASE

Abstract

The occurrence of diseases that impact the leaves of corn plants presents a substantial obstacle in agriculture, leading to a reduction in the overall yield of crops. This study aims to perform a comparative analysis of transfer learning methodologies by employing three distinct ResNet architectures: ResNet18, ResNet50, and ResNet101. The dataset utilized by the author consists of a compilation of images portraying corn leaves that demonstrate varying levels of disease severity. Transfer learning refers to leveraging a pre-existing ResNet model and retraining the network by employing the corn leaf dataset. The experimental results demonstrate that the ResNet18, ResNet50, and ResNet101 models achieved accuracy rates of 96.68%, 95.73%, and 95.26%, respectively. The ResNet101 model shows superior performance in terms of precision and recall metrics. This research indicates that utilizing a more complex and sophisticated network structure can improve the effectiveness of disease identification in corn plant leaves. The result above is essential in promoting sustainable agricultural methodologies and efficiently managing corn plant diseases.