改进残差神经网络在遥感图像分类中的应用

针对传统卷积神经网络随着深度加深而导致网络退化以及计算量大等问题,提出一种改进残差神经网络的遥感图像场景分类方法。该方法以残差网络ResNet50作为主框架,在残差结构中引入深度可分离卷积和分组卷积,减少了网络的参数量和计算量,加快模型收敛的同时也提升了分类精度。此外在网络中嵌入多尺度SE block模块对通道特征进行重校准,提取出更加重要的特征信息,进一步提升了网络的分类性能。在AID和UCMerced_Land Use两个公开数据集上的分类精度分别为91.92%和93.52%,相比常规残差网络分类精度分别提高了3.38%和10.24%,证明所提方法在遥感图像场景分类任务中的可行性和有效性。

Application of Improved Residual Network in Sensing Image Classification

In view of the degradation of traditional convolution neural network with the deepening of depth and the large amount of calculation, a remote sensing image scene classification method based on improved residual neural network is proposed.This method takes the residual network ResNet50 as the main framework, and introduces deep separable convolution and grouping convolution into the residual structure, which reduces the amount of network parameters and calculation, speeds up the convergence of the model and improves the classification accuracy. In addition, the multi-scale SE block module is embedded in the network to recalibrate the channel features and extract more im-portant feature information, which further improves the classification performance of the network. The classification accuracy of AID and UCMerced_Land Use is 91.92% and 93.52% respectively, which is 3.38% and 10.24% higher than that of conventional residual network, which proves the feasibility and effectiveness of the proposed method in remote sensing image scene classification task.