Search

The accurate categorization of crop kinds with hyperspectral remote sensing imaging (HRSI) represents an imperative adaption in agriculture and is of huge emergence in estimating the yield of crops and monitoring growth. Amidst the deep models, convolutional neural networks (CNNs) tend to be the best technique for hyperspectral image (HSI) classification. However, the classical CNN poses a fixed shape, which poses simulation complexity in modeling huge ranges. To handle the problem, a two-level classification technique is developed for crop classification using HSI. The aim is to design a Spinal Net-assisted multi-level crop classification using HSI. Here, a hyperspectral satellite crop image is taken as input, and the anisotropic diffusion is used for pre-processing. U-Net is applied to segment crops. The feature mining is executed considering spectral-spatial features, and vegetation index features. The hyperspectral crop identification process is engaged with Squeeze Net, which is trained by fractional light spectrum optimizer (FLSO). In this step, if the hyperspectral crop is identified, then the multi-level crop type classification process is carried out using FLSO-based Spinal Net. The FLSO-based Spinal Net provided better accuracy of 92.2%, false positive rate (FPR) of 0.083, and true positive rate (TPR) of 91.7%.