论文部分内容阅读
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 characteristics of the image, the two groups of BP neural networks are used to classify the features. One group uses the features extracted from the true color image as the network input and the other group uses the transformed gray image as the feature input. Complete the task of coarse and subdivision of image.On the basis of experiment, we put forward the principle of selecting proportion of all kinds of samples, that is to say, taking the ratio of the number of all kinds of objects in original image as reference, on the basis of which, The selection of the eigenvectors uses the rectangular area with the same proportion as the image features as the selection range.It uses the median filtering technique to filter out the extra eigen elements in the image preprocessing and sharpens the eigenvector information.With the subsequent processing of the result , Used to filter out noise points, the classification results more clearly.