基于BP神经网络和中值滤波的地质雷达图像分类

Classifying the stratums in the images from the geology radar sonde based on the BP neutral network and median filtering

针对地质冒达图像的特点,使用两组BP神经网络进行分类,一组使用从真彩色图像中提取的特征作为网络输入,另一组使用转化后的灰度图像作为特征输入,两组网络分别完成图像的粗分与细分的任务.在实验基础上提出各类样本个数的选取比例原则,即以原图中各类地物个数的比值作为参考,在此基础上做适当的增减.特征向量的选取使用了与图像特征比例一致的长方形区域作为选择范围.使用中值滤波技术,在图像预处理中用于滤除多余特征元素,锐化特征信息;在对结果的后续处理中,用于滤除噪声点,使分类结果更加清晰.

According to the trait of the radar image,two group of BP neutral network are used to classify the stratum of the image.In one group,the traits of the radar image are extracted from the real color image as the input of the network,while in another one,the traits are extracted from the grey image.The two groups singly accomplish the tasks of rude classifying and meticulous classifying.In the use of the neutral network,in the select of the samples,the select principle of the ratio of the number of different classes is proposed based on the experiments.The ratio should be based on the ratio of the number of different classes in the original images,and proper change may be available.In the select of the trait vector,according to the practice of the project,a rectangle district whose shape is similar to the trait of the image is selected as the trait vector of the network,and better result is acquired.The median filtering technology is used.In the pretreatment of the image,it is used to strain the superabundance trait element and intensify the infor- mation of the traits;and in the disposal of the result,it is used to strain the yawp point,and make the re- sult more clear.