Image processing method and device, and storage medium

1. An image processing method, wherein the method comprises:

performing feature extraction on an image to be processed to obtain a first feature map of the image to be processed;

splitting the first feature map into a plurality of first sub-feature maps according to dimension information of the first feature map and a preset splitting rule, the dimension information of the first feature map comprising dimensions of the first feature map and size of each dimension;

performing normalization on the plurality of first sub-feature maps respectively to obtain a plurality of second sub-feature maps;

splicing the plurality of second sub-feature maps to obtain a second feature map of the image to be processed;

encoding the second feature map of the image to be processed to obtain a plurality of encoded feature maps, each of the encoded feature maps having different scales; and

decoding the plurality of encoded feature maps to obtain a classification prediction result of the image to be processed.

2. The method according to claim 1 , wherein splitting the first feature map into the plurality of first sub-feature maps according to the dimension information of the first feature map and the preset splitting rule comprises:

according to sizes of spatial dimensions of the first feature map and the preset splitting rule, splitting the first feature map in the spatial dimensions to obtain the plurality of first sub-feature maps.

3. The method according to claim 1 , wherein performing normalization on the plurality of first sub-feature maps respectively to obtain the plurality of second sub-feature maps comprises:

grouping each first sub-feature map in a channel dimension, and performing normalization on each group of channels of the first sub-feature map to obtain a second sub-feature map of the first sub-feature map.

4. The method according to claim 1 , wherein splicing the plurality of second sub-feature maps to obtain the second feature map of the image to be processed comprises:

according to positions of the plurality of first sub-feature maps in the first feature map, splicing the plurality of second sub-feature maps to obtain the second feature map of the image to be processed.

5. The method according to claim 1 , wherein the splitting rule comprises at least one of the following: dimensions to be split of the feature map, a splitting position of each dimension to be split, a splitting number of each dimension to be split, a splitting size of each dimension to be split, and a number of split sub-feature maps.

6. The method according to claim 1 ,

wherein the encoding the second feature map of the image to be processed to obtain a plurality of encoded feature maps further comprises performing scale reduction and multi-scale fusion on the second feature map of the image to be processed to obtain the plurality of encoded feature maps; and

wherein the decoding the plurality of encoded feature maps to obtain a classification prediction result of the image to be processed further comprises performing scale amplification and multi-scale fusion on the plurality of encoded feature maps to obtain the classification prediction result of the image to be processed.

7. The method according to claim 6 , wherein performing scale reduction processing and multi-scale fusion processing on the at least one second feature map of the image to be processed to obtain the plurality of encoded feature maps comprises:

performing scale reduction on m second feature maps to obtain m scaled-down feature maps, m being a positive integer;

performing feature fusion on the m scaled-down feature maps to obtain an (m+1)th feature map, scales of the m scaled-down feature maps being equal to that of the (m+1)th feature map; and

performing feature optimization and fusion on the m second feature maps and the (m+1)th feature map respectively to obtain m+1 encoded feature maps.

8. The method according to claim 7 , wherein performing scale amplification processing and multi-scale fusion processing on the plurality of encoded feature maps to obtain the classification prediction result of the image to be processed comprises:

performing fusion and scale amplification on the m+1 encoded feature maps to obtain m scaled-up feature maps, m being a positive integer; and

performing feature optimization and fusion on the m scaled-up feature maps to obtain the classification prediction result of the image to be processed.

9. The method according to claim 1 , wherein the method is implemented by a neural network, which comprises a feature extraction network for performing feature extraction on the image to be processed, an encoding network for performing scale reduction and multi-scale fusion on at least one second feature map of the image to be processed, and a decoding network for performing scale amplification and multi-scale fusion on a plurality of encoded feature maps.

10. The method according to claim 9 , wherein the method further comprises:

training the neural network according to a preset training set, the training set comprising a plurality of labeled sample images.

11. An image processing device, comprising:

a processor; and

a memory configured to store processor-executable instructions,

wherein the processor is configured to invoke the instructions stored in the memory to perform operations comprising:

performing feature extraction on an image to be processed to obtain a first feature map of the image to be processed;

splitting the first feature map into a plurality of first sub-feature maps according to dimension information of the first feature map and a preset splitting rule, the dimension information of the first feature map comprising dimensions of the first feature map and size of each dimension;

performing normalization on the plurality of first sub-feature maps respectively to obtain a plurality of second sub-feature maps;

splicing the plurality of second sub-feature maps to obtain a second feature map of the image to be processed;

encoding the second feature map of the image to be processed to obtain a plurality of encoded feature maps, each of the encoded feature maps having different scales; and

decoding the plurality of encoded feature maps to obtain a classification prediction result of the image to be processed.

12. The device according to claim 11 , wherein splitting the first feature map into the plurality of first sub-feature maps according to the dimension information of the first feature map and the preset splitting rule comprises:

splitting the first feature map in spatial dimensions according to sizes of the spatial dimensions of the first feature map and a preset splitting rule to obtain the plurality of first sub-feature maps.

13. The device according to claim 11 , wherein performing normalization on the plurality of first sub-feature maps respectively to obtain the plurality of second sub-feature maps comprises:

grouping each first sub-feature map in a channel dimension, and perform normalization on each group of channels of the first sub-feature map respectively to obtain a second sub-feature map of the first sub-feature map.

14. The device according to claim 11 , wherein splicing the plurality of second sub-feature maps to obtain the second feature map of the image to be processed comprises:

splicing the plurality of second sub-feature maps according to positions of the plurality of first sub-feature maps in the first feature map to obtain the second feature map of the image to be processed.

15. The device according to claim 11 , wherein the splitting rule comprises at least one of the following: dimensions to be split of the feature map, a splitting position of each dimension to be split, a splitting number of each dimension to be split, a splitting size of each dimension to be split, and a number of split sub-feature maps.

16. The device according to claim 11 ,

wherein the encoding the second feature map of the image to be processed to obtain a plurality of encoded feature maps further comprises performing scale reduction and multi-scale fusion on the second feature map of the image to be processed to obtain the plurality of encoded feature maps; and

wherein the decoding the plurality of encoded feature maps to obtain a classification prediction result of the image to be processed further comprises performing scale amplification and multi-scale fusion on the plurality of encoded feature maps to obtain the classification prediction result of the image to be processed.

17. The device according to claim 16 , wherein performing scale reduction and multi-scale fusion on the at least one second feature map of the image to be processed to obtain the plurality of encoded feature maps comprises:

performing scale reduction on m second feature maps to obtain m scaled-down feature maps, m being a positive integer;

performing feature fusion on the m scaled-down feature maps to obtain an (m+1)th feature map, scales of the m scaled-down feature maps being equal to that of the (m+1)th feature map; and

performing feature optimization and fusion on the m second feature maps and the (m+1)th feature map respectively to obtain m+1 encoded feature maps.

18. The device according to claim 17 , wherein performing scale amplification and multi-scale fusion on the plurality of encoded feature maps to obtain the classification prediction result of the image to be processed comprises:

performing fusion and scale amplification on the m+1 encoded feature maps to obtain m scaled-up feature maps, m being a positive integer; and

performing feature optimization and fusion on the m scaled-up feature maps to obtain the classification prediction result of the image to be processed.

19. The device according to claim 11 , wherein the instructions are invoked to run a neural network so as to perform the steps.

20. A non-transitory computer readable storage medium, having computer program instructions stored thereon, wherein when the computer program instructions are executed by a processor, the processor is caused to perform operations of:

performing feature extraction on an image to be processed to obtain a first feature map of the image to be processed;

splitting the first feature map into a plurality of first sub-feature maps according to dimension information of the first feature map and a preset splitting rule, the dimension information of the first feature map comprising dimensions of the first feature map and size of each dimension;

performing normalization on the plurality of first sub-feature maps respectively to obtain a plurality of second sub-feature maps;

splicing the plurality of second sub-feature maps to obtain a second feature map of the image to be processed;

encoding the second feature map of the image to be processed to obtain a plurality of encoded feature maps, each of the encoded feature maps having different scales; and

decoding the plurality of encoded feature maps to obtain a classification prediction result of the image to be processed.