Benchmarking CNN-Based Systems for Corn Leaf Pest Detection using Fine-Tuning

This research presents a computer vision system for the detection of diseases in maize leaves using convolutional neural networks. The Peruvian valley of Chicama was the focus of our study, where images were collected and subsequently added to the Plant Village dataset. Image preprocessing techniques, including GrabCut and data augmentation, were employed to enhance the quality of the images. The objective of this research was to compare a number of fine-tuned architectures, including DenseNet121, DenseNet201, ResNet50, ResNet101, VGG16 and VGG19, in order to identify the most suitable model for the classification of maize leaf diseases.The results demonstrated that VGG16 achieved the highest accuracy of 93.16%. DenseNet121 followed closely with an accuracy of 93.03%, indicating its strong performance. In contrast, ResNet50 showed the lowest accuracy at 87.94%.