基于多尺度半耦合卷积稀疏编码的遥感地貌影像纹理识别方法

Texture recognition method of remote sensing landform image based on multi-scale semi-coupled convolutional sparse coding

摘要:

遥感地貌影像通常包含大量的数据,具有高度的复杂性和多样性,难以捕捉到不同层次的纹理信息,从而影响识别效果。因此,为提高纹理特征提取的效果,确保识别精度,提出基于多尺度半耦合卷积稀疏编码的遥感地貌影像纹理识别方法研究。去除遥感地貌影像噪声,增强遥感地貌影像整体质量,通过分水岭算法分割遥感地貌影像,探究不同尺度下遥感地貌影像纹理特征区别。然后应用灰度共生矩阵(GLCM)获取遥感地貌影像的多尺度纹理特征,构建半耦合卷积稀疏编码模型,完成多尺度纹理特征提取过程的学习与多尺度纹理特征的有效融合,并选取适当的分类器——朴素贝叶斯分类器,并对其进行训练。最后以此为基础,制定遥感地貌影像纹理识别程序,执行制定程序即可获取地貌纹理识别结果。测试结果显示:应用提出方法获得的遥感地貌影像处理结果清晰度与对比度较高,地貌纹理特征提取结果更加完整与清晰,地貌纹理识别结果与实际结果一致,充分证实了提出方法应用效果更好。

Abstract:

Remote sensing landform images usually contain a large amount of data, which is highly complex and diverse, making it difficult to capture texture information at different levels, thereby affecting recognition performance. Therefore, in order to improve the effectiveness of texture feature extraction and ensure recognition accuracy, a texture recognition method for remote sensing topographic images based on multi-scale semi coupled convolutional sparse encoding is proposed. Remove noise from remote sensing landform images, enhance the overall quality of remote sensing landform images, segment remote sensing landform images through watershed algorithm, and explore the differences in texture features of remote sensing landform images at different scales. Then, the gray level co-occurrence matrix (GLCM) is applied to obtain multi-scale texture features of remote sensing geomorphic images, and a semi coupled convolutional sparse encoding model is constructed to complete the learning of multi-scale texture feature extraction process and effective fusion of multi-scale texture features. An appropriate classifier - Naive Bayes classifier is selected and trained. Finally, based on this, a remote sensing terrain image texture recognition program is developed, and the results of terrain texture recognition can be obtained by executing the program. The test results show that the remote sensing landform image processing results obtained by the proposed method have high clarity and contrast, and the terrain texture feature extraction results are more complete and clear. The terrain texture recognition results are consistent with the actual results, fully confirming that the proposed method has better application effect.