Deep Residual Learning for Hyperspectral Imaging Camouflage Detection with SPXY-Optimized Feature Fusion Framework

Camouflage detection in hyperspectral images remains challenging due to the high spectral similarity between artificial materials and natural vegetation. This study aimed to develop a non-destructive classification framework for distinguishing camouflage fabrics from natural grass based on reflectance data within the 389.06-1005.10 nm range. The workflow involved Principal Component Analysis (PCA) performed using ENVI software for dimensionality reduction, followed by Standard Normal Variate (SNV) transformation, Savitzky–Golay(SG) filtering, first-order derivative processing, and Z-score normalization. The Sample set Partitioning based on joint X-Y distance (SPXY) algorithm was utilized to enhance sample representativeness. Six classification models were constructed, including PCA, Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), Convolutional Neural Networks (CNNs), and Residual Neural Network (ResNet). The SPXY-ResNet model achieved the highest classification performance, with an accuracy of 99.17%, a precision of 98.89%, a recall of 98.82%, and an F1-score of 98.82%. Model calibration under different sampling strategies was evaluated using Kullback-Leibler (KL) divergence, mean similarity, and standard deviation of similarity. Results indicated that the SPXY algorithm improved distributional consistency between training and testing sets, thereby enhancing generalization performance. The confusion matrix revealed only two misclassifications among 120 test samples, indicating strong classification reliability. Convergence curves further confirmed stable training behavior, with no signs of overfitting. These results demonstrate that the proposed SPXY-ResNet framework offers accurate and robust performance for camouflage detection in complex spectral environments.