A novel approach using hybrid deep features for citrus disease detection and classification based on NCA and Bayesian optimised random forest classifier

In this paper, deep features are retrieved from the citrus disease image gallery dataset using the AlexNet convolutional neural network. After data augmentation, the image dataset is sent for data pre-processing, where colour moments, GLCM, and Gabor wavelets are used to extract the colour feature and texture. Hybrid features are created by combining colour, texture, and deep characteristics. For feature selection, neighbourhood component analysis (NCA) is employed. The model's classification capabilities were also evaluated with those of the random forest classifier architecture for citrus disease detection in case 1 images and the Bayesian optimised random forest classifier architecture for citrus fruit disease detection in case 2 images. On a benchmark dataset including images of four distinct illnesses and healthy citrus fruit classes, the model performances were discovered. The dataset was identified using the disease image dataset with a maximum accuracy rate of 95.07%.