Deep Learning-Augmented Hybrid Optimization for Crop Classification using Hyperspectral images

Precise and efficient solutions are necessary for the essential job of classifying crops using hyperspectral images in precision agriculture. The paper introduces a framework called Deep Learning-Augmented Hybrid Optimization (DLAHO), which combines a CNN for feature extraction with a hybrid optimization technique that combines Genetic Algorithms (GA) and Gradient-Based Optimization (GBO) to optimize model parameters. The Hyperspectral Image Database (HID) preprocesses hyperspectral images by using normalization and PCA to reduce dimensionality. The CNN utilizes both spectral and spatial data, which are then enhanced by the hybrid GA-GBO method. The improved model accurately identifies crops with a high level of precision. The results indicate that DLAHO obtains an accuracy of 95.4% for wheat, 92.1% for maize, 94.3% for soybean, 93.2% for barley, 96.0% for rice, 91.5% for sugarcane, 93.8% for potato, and 92.7% for cotton. DLAHO outperforms baseline models and state-of-the-art algorithms, achieving an F1-score of 93.1%, recall of 93.5%, precision of 92.8%, and total accuracy of 93.6%. The findings show that the recommended DLAHO architecture significantly increases the precision and dependability of crop categorization, thereby making it a viable tool for precision agriculture.