Cotton crop classification using satellite images with score level fusion based hybrid model

Abstract

Accurate cotton images are significant component for surveiling cotton development and its precise control. A suitable technique for charting the distribution of cotton at the county or field level must be available to researchers and production managers. The classification of cotton remote sensing models at the county level has significant implications for precision farming, land management, and government decision-making. This work aims to develop a novel cotton crop classification model using satellite images based on soil behaviour. It includes phases like preprocessing, segmentation, feature extraction, and classification. Here, preprocessing is carried out by Gaussian filtering to improve the quality of the input image. Then Modified Deep Joint Segmentation method is employed for the segmentation process. The features such as wide dynamic range vegetation index, simple ratio, Green Chlorophyll index, Transformed vegetation index, and Green leaf area index are extracted for classifying the input. The hybrid Improved CNN (ICNN) and Bidirectional Gated recurrent Unit (Bi-GRU) have used for classification purposes, which is computed by the improved score level fusion. The suggested new hybrid optimization model known as the Battle Royale assisted Butterfly optimization algorithm (BRABOA) is used for adjusting the hidden neuron count of both the ICNN and Bi-GRU classifiers for improving the accuracy. At last, the efficiency of the suggested model is then evaluated to other schemes using a variety of metrics. The suggested HC + BRABOA method obtains a maximum accuracy of (0.95) over conventional methods at a learning percentage of 90% for classifying cotton crops using satellite images.