Sustainable Agriculture-Based Climate Change Training Models using Remote Hyperspectral Image with Machine Learning Model

• Published in: Remote sensing in earth systems sciences (Online) Vol. 7; no. 4; pp. 261 - 270
• Main Authors: Durairaj, M., Rubenraju, Kasapaka, Krishna, B. V. Rama, Yajid, Mohd Shukri Ab, Jackson, Beulah, Babu, Jampani Satish, Padmavathi, Kodali Lakshmi
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.12.2024
• Subjects: Earth and Environmental Science; Earth System Sciences; Geography; Monitoring/Environmental Analysis; Remote Sensing/Photogrammetry; Climate change; Hyperspectral image analysis; Sustainable agriculture; Machine learning; Classification
• ISSN: 2520-8195; 2520-8209
• DOI: 10.1007/s41976-024-00118-y

In order to help farmers and crop managers better understand the elements influencing crop status and growth, hyperspectral and multispectral data processing methods have shown to be beneficial. Utilising advanced computational methods via machine learning is one strategy that has been in use recently. This method can forecast satellite image data based on the circumstances of mapping different types of land and vegetation in the field. This research proposes novel technique in sustainable agriculture-based climate change detection using hyperspectral image analysis with machine learning model. Here, the hyperspectral image of agricultural field is collected as input and processed for smoothening with normalisation. The proposed image analysis model is carried out in two stages which is feature extraction and classification. In stage 1, the feature extraction of processed input hyperspectral image is carried out using multilayer Bayesian encoder vector model (MBEV). The second stage of this proposed model is to classify the extracted image using deep convolutional belief neural networks (DCBNN). The experimental analysis has been carried out for various agriculture-based hyperspectral image datasets in terms of training accuracy, sensitivity, specificity, and AUC. The experimental findings demonstrate that, when compared to other ways, the suggested strategy performed exceptionally well. Proposed technique attained training accuracy of 97%, AUC of 85%, sensitivity of 96%, and specificity of 93%.