Design of a Fast Feature Mining Algorithm for Sensor Images
To solve the problem of image feature loss due to exposure during sensor image acquisition,which leads to high feature error mining rate and poor PR curve performance,a new fast sensor image feature mining algorithm is proposed. PReLU function is used as the activation function to integrate the parametric rectified linear unit and input images. The loss function and the SSIM loss function are used to constrain the image in the training stage. Softmax regression is used to mine the image features,and the convolutional neural net-work sensor image feature mining structure is constructed. AdaGrad is used to update the structural parameters,and mining steps are de-signed to realize fast mining of sensor image features. The experimental results show that the designed convolutional neural network im-age feature mining method has higher PSNR value,the SSIM value is closer to 1,the error mining rate is 1.0％,the PR curve is closest to the upper right corner,the area is the largest,the image visual effect is better,and the maximum feature mining time is only 1.3 s, which realizes fast mining.
万方数据
维普
