Classification of Corn Leaf Images using Pre-Trained Deep Learning Schemes: A Study

Automatic image classification using pre-trained deep learning (PDL) schemes are widely employed in several domains. This research aims to verify the classification performance of the chosen PDL schemes using the Corn Leaf Image (CLI) dataset. Stages existing in the proposed system include; (i) CLI collection, resizing and pre-processing with histogram equalization, (ii) feature extraction using chosen PDL scheme, (iii) SoftMax supported binary classification, and (iv) performance evaluation and verification. In this study, 3000 CLI images (1500 normal and 1500 rust instances) are examined. The experimental investigation of the study reveals that the VGG16 scheme exhibits superior classification accuracy on the chosen CLI dataset. Specifically, it achieves an accuracy of >93% when considering raw images and attains 100% accuracy when using pre-processed images. This research confirms that, in comparison to other models such as VGG19, ResNet variants, and DenseNet variants investigated in this study, VGG16 provides better result for the CLI dataset.