Noise reduction auto-encoder-based anomaly detection model training method
View Patent ↗Disclosed is a method for training an abnormal-detection model based on an improved denoising autoencoder, including: acquiring an original image; generating a rectangular frame according to a preset range of a resolution ratio, by the improved denoising autoencoder, and occluding the original image by the rectangular frame, wherein the resolution ratio is a ratio of a resolution of an occlusion area formed by the rectangular frame to a resolution of the original image; filling random noise in the rectangular frame to obtain a noised image, by the improved denoising autoencoder; and performing constraint learning on the original image and the noised image, by the abnormal-detection model, to implement training of the abnormal-detection model. Because a learning task is more complex, it is helpful to alleviate identity mapping, and detection performance of the model is improved. The present application further provides an apparatus, a device, and a readable storage medium thereof.
1. A method for training an abnormal-detection model based on an improved denoising autoencoder, comprising:
acquiring an original image;
generating a rectangular frame according to a preset range of a resolution ratio, by the improved denoising autoencoder, and occluding the original image by the rectangular frame, wherein the resolution ratio is a ratio of a resolution of an occlusion area formed by the rectangular frame to a resolution of the original image;
filling random noise in the rectangular frame to obtain a noised image, by the improved denoising autoencoder; and
performing constraint learning on the original image and the noised image, by the abnormal-detection model, to implement training of the abnormal-detection model.
2. The method according to claim 1 , wherein the generating the rectangular frame according to the preset range of the resolution ratio, by the improved denoising autoencoder, comprises:
randomly generating a target resolution ratio within the preset range of the resolution ratio; and
generating the rectangular frame according to the target resolution ratio, by the denoising autoencoder.
3. The method according to claim 1 , wherein the generating the rectangular frame according to the preset range of the resolution ratio, by the improved denoising autoencoder, comprises:
generating a random number of the rectangular frames according to the preset range of the resolution ratio, by the improved denoising autoencoder.
4. The method according to claim 1 , wherein the generating the rectangular frame according to the preset range of the resolution ratio, by the improved denoising autoencoder, comprises:
within a preset range of a length-width ratio of the rectangular frame, randomly generating a target length-width ratio; and
generating the rectangular frame whose length-width ratio is the target length-width ratio, according to the preset range of the resolution ratio, by the improved denoising autoencoder.
5. The method of claim 1 , wherein the occluding the original image by the rectangular frame, comprises:
randomly determining a position coordinate of the rectangular frame in the original image; and
occluding the original image according to the position coordinate, by the rectangular frame.
6. The method according to claim 1 , wherein the filling the random noise in the rectangular frame by the improved denoising autoencoder, comprises:
filling the random noise in the rectangular frame according to a preset probability density of the random noise, by the improved denoising autoencoder.
7. The method of claim 1 , wherein
the random noise comprises any one item or more items of following: Gaussian noise, salt-and-pepper noise, Poisson noise, and Laplace noise.
8. A device for training an abnormal-detection model based on an improved denoising autoencoder, comprising:
a storage, configured to store a computer program; and
a processor, configured to execute the computer program, to implement the method for training the abnormal-detection model based on the improved denoising autoencoder according to claim 1 .
9. A non-transitory readable storage medium, on which a computer program is stored, wherein, in response to that the computer program is executed by a processor, the processor implements the method for training the abnormal-detection model based on the improved denoising autoencoder according to claims 1 .
10. The method according to claim 1 , wherein the resolution is a display resolution.
11. The method according to claim 1 , wherein the ratio of the resolution of the occlusion area formed by the rectangular frame to the resolution of the original image equals a ratio of an area of the occlusion area formed by the rectangular frame to an area of the original image.
12. The method according to claim 4 , wherein the length-width ratios of the respective rectangular frames generated during different generation processes of the rectangular frames are different.
13. The method according to claim 4 , wherein the length-width ratios of the respective rectangular frames generated during a same generation process of the rectangular frame are different.
14. The method according to claim 4 , wherein the length-width ratios of the respective rectangular frames during a same generation process of the rectangular frame are the same, and the length-width ratios of the respective rectangular frames during different generation processes of the rectangular frames are different.
15. The method according to claim 1 , wherein the rectangular frame is determined according to one or more of random manners, and the random manners comprise:
the resolution ratio is random;
a number of the rectangular frame is random;
a length-width ratio of the rectangular frame is random; and
a position coordinate of the rectangular frame in the original image is random.
16. The method according to claim 1 , wherein pixel points in the rectangular frame are set to black, white or gray.
17. The method according to claim 7 , wherein, in response to that the random noise comprises two items or more items of following: the Gaussian noise, the salt-and-pepper noise, the Poisson noise, and the Laplace noise, the two items or more items are superimposed in the rectangular frame.
18. The method according to claim 1 , wherein the original image is selected from normal images or abnormal images.
19. The method according to claim 1 , wherein the preset range is 10%-20%.
20. The method according to claim 1 , wherein the rectangular frame acts as a local occlusion to force the abnormal-detection model to comprehensively learn global information of the noised image, and to make the abnormal-detection model stereotypically learn a denoising method in a patterned way during a learning process.