融合MF-DWPSO-BP神经网络的遥感影像作物分类方法OA

To address the issues of single-feature reliance,susceptibility to local optima,slow convergence,and weak generalization in traditional remote sensing image-based crop classification,this study proposes a Back Propagation(BP)neural network method integrated with Multi-dimensional Features(MF)and a Dynamic Weight Particle Swarm Optimization(DWPSO)algorithm.A"spectral-texture"feature system was constructed by extracting spectral features,such as the Normalized Difference Vegetation Index(NDVI)and Enhanced Vegetation Index(EVI)of crops,and combining them with texturefeatures,includingenergyand entropy,derived from the Gray-Level Co-occurrence Matrix(GLCM).The dynamic weight particle swarm optimization algorithm was designed to optimize the parameters of the BP neural network.Crop classification was then performed using the optimized BP neural network,followed by a post-processing step involving small patch removal to enhance practical utility.Experiments were conducted using Landsat-8(30m)and Sentinel-2(10m)agricultural images,with comparisons made against traditional BP and fixed-weight PSO-BP algorithms.The results demonstrated that the MF-DWPSO-BP method achieved an overall accuracy(OA)of 93.0% and a Kappa coefficient of 0.91 on Landsat-8 imagery,which were 6.8 and 8 percentage points higher,respectively,than those of the fixed-weight PSO-BP algorithm.On Sentinel-2 imagery,it achieved an OA of 92.0% and a Kappa coefficient of 0.90.Moreover,the proposed method achieved convergence with 50% fewer iterations compared to the traditional BP network.This method can improve the classification accuracy for spectrally similar crops and provides support for agricultural monitoring.