基于分形维度的林业遥感图像树种分类识别

Classification and Recognition of Tree Species in Forestry Remote Sensing Image Based on Fractal Dimension

传统的树种分类识别方法未进行最大池化操作,导致树种分类识别精度差。现引入分形维度进行林业遥感图像树种分类识别。通过ROI区域截取获取遥感树种图像,利用直方图均衡化方法进行原始图像预处理,以便获得高质量与清晰度的林业遥感图像;通过分形维度理论分析提取的林业遥感图像纹理特征,完成卷积神经网络模型的优化构建;将林业遥感图像纹理特征输入卷积层,经卷积层的卷积操作并计算特征数据,池化池通过最大池化操作卷积层输出的数据;通过Relu激活函数对林业遥感图像树种纹理特征进行深度分析,利用Softmax分类器实现树种分类识别。实验结果表明,上述方法预处理后的遥感图像质量高,且林业遥感图像树种分类识别的效率高,分类识别的时间低至35.7ms,分类识别的准确率高达95.62%。

The traditional classification and recognition methods of tree species do not carry out maximum pooling operation, which leads to poor classification and recognition accuracy of tree species. Therefore, fractal dimension was introduced to classify and recognize tree species in forestry remote sensing images. In order to obtain high quality and clear forest remote sensing image, original images were preprocessed by histogram equalization method by capturing remote sensing tree images in ROI region. The fractal dimension theory was used to analyze the extracted forest remote sensing image texture features, and the convolutional neural network model was optimized. The texture features of forestry remote sensing images were input into the convolution layer, and the characteristic data were calculated after the convolution operation of the convolution layer. The pooled pool output the data of the convolution layer through the maximum pooling operation. The texture features of tree species in forest remote sensing images were deeply analyzed by ReLU activation function, and the classification and recognition of tree species were realized by Softmax classifier. The experimental results show that the quality of the remote sensing images preprocessed by this method is high, and the classification and recognition efficiency of forest remote sensing image tree species is high, the classification and recognition time is as low as 35.7 ms, and the classification and recognition accuracy is as high as 95.62%.