Image processing method, computer system, electronic device, and program product
Image processing is described. An example method includes generating a feature extraction layer portion of an image classification model based on a deep neural network (DNN) model, and extracting features of a group of images by using the feature extraction layer portion. The method further includes training an output layer portion of the image classification model according to the features of training images in the group of images and classification labels of the training images. The method further includes generating the image classification model by combining the feature extraction layer portion and the output layer portion. Embodiments of the present disclosure implement a lightweight artificial intelligence (AI) solution on a storage system. An expanded storage system can facilitate the generation of AI applications and assist in training an image classification model more quickly, and the obtained image classification model can also produce high accuracy on a small training set.
1. A method, comprising:
training, by a system comprising a processor, a first deep neural network (DNN) model for image classification of a group of categories of images, wherein the first DNN model comprises a feature extraction layer portion and a first output layer portion, and wherein the feature extraction layer portion only comprises an input layer and one or more hidden layers, and does not comprise any output layers of the first DNN model;
extracting, by the system, the feature extraction layer portion from the first DNN model;
incorporating, by the system, the feature extraction layer portion as an executable module in a storage device of the system;
extracting, by the system, using the feature extraction layer portion in the storage device, respective features of images stored in the storage device;
storing, by the system, in the storage device, the respective features of the images;
retrieving, by the system, from the storage device, the respective features of training images in the images, wherein the training images are associated with a sub-group of the group of categories of images;
training, by the system, a second output layer portion for image classification of the sub-group of the group of categories of images, based on the respective features of the training images and according to classification labels of the training images, wherein the training of the second output layer portion does not comprise training the feature extraction layer portion; and
generating, by the system, a second DNN model for image classification of the sub-group of the group of categories of images by combining the feature extraction layer portion and the second output layer portion.
2. The method according to claim 1 , wherein the first DNN model is trained using a subset of the images.
3. The method according to claim 1 , further comprising:
receiving, by the system, a new image; and
classifying, by the system, using the second DNN model, the new image.
4. The method according to claim 1 , further comprising:
receiving, by the system, additional training images; and
extracting, by the system, using the feature extraction layer portion, respective features of the additional training images,
wherein the training of the second output layer portion comprises training the second output layer portion based further on the respective features of the additional training images.
5. The method according to claim 1 , further comprising:
adjusting, by the system, the feature extraction layer portion and the second output layer portion of the second DNN model based on the training images and the classification labels of the training images.
6. The method according to claim 5 , wherein adjusting the feature extraction layer portion and the output layer portion of the second DNN model comprises:
training the second DNN model by using original data of the training images and the classification labels.
7. A system, comprising:
at least one storage device;
at least one processing unit; and
at least one memory that is coupled to the at least one processing unit and stores instructions for execution by the at least one processing unit, the at least one processing unit comprising:
an image classification model generation unit, configured to train a first deep neural network (DNN) model for image classification of a group of categories of images, wherein the first DNN model comprises a feature extraction layer portion and a first output layer portion, and wherein the feature extraction layer portion comprises an input layer and one or more hidden layers, and does not comprise any output layers of the first DNN model; and
a feature extraction unit configured to:
extract the feature extraction layer portion from the first DNN model;
incorporate the feature extraction layer portion as an executable module in the at least one storage device of the system; and
extract, using the feature extraction layer portion in the at least one storage device, respective features of images stored in the at least one storage device;
storing, in the at least one storage device, the respective features of the images; and
wherein image classification model generation unit is further configured to:
retrieve, from the storage device, the respective features of training images in the images, wherein the training images are associated with a sub-group of the group of categories of images;
train a second output layer portion for image classification of the sub-group of the group of categories of images, based on the respective features of the training images and according to classification labels of the training images, wherein the training of the second output layer portion does not comprise training the feature extraction layer portion; and
generate a second DNN model for image classification of the sub-group of the group of categories of images by combining the feature extraction layer portion and the second output layer portion.
8. The system according to claim 7 , wherein the first DNN model is trained using a subset of the images.
9. The system according to claim 7 , wherein the image classification model generation unit is further configured to:
receive a new image; and
classify, using the second DNN model, the new image.
10. The system according to claim 7 , wherein the image classification model generation unit is further configured to:
receive additional training images;
extract, using the feature extraction layer portion, respective features of the additional training images; and
train the second output layer portion based further on the respective features of the additional training images.
11. The system according to claim 7 , wherein the image classification model generation unit is further configured to further adjust the feature extraction layer portion and the second output layer portion of the second DNN model based on the training images and the classification labels of the training images.
12. The system according to claim 11 , wherein the image classification model generation unit is further configured to train the second DNN model by using original data of the training images and the classification labels.
13. The system according to claim 11 , wherein the image classification model is further configured to adjust the feature extraction layer portion and the output layer portion of the second DNN model by using at least one graphics processing unit.
14. A non-transitory computer-readable medium, comprising machine-executable instructions that, in response to execution by a system comprising a processor, cause the device to perform operations, comprising:
training a first deep neural network (DNN) model for image classification of a group of categories of images, wherein the first DNN model comprises a feature extraction layer portion and a first output layer portion, and wherein the feature extraction layer portion comprises an input layer and one or more hidden layers, and does not comprise any output layers of the first DNN model;
extracting the feature extraction layer portion from the first DNN model;
incorporating the feature extraction layer portion as an executable module in a storage device of the system;
extracting, using the feature extraction layer portion in the storage device, a group of features of respective features of images stored in the storage device;
storing, in the storage device, the respective features of the images;
retrieving, from the storage device, the respective features of training images in the images, wherein the training images are associated with a sub-group of the group of categories of images;
training a second output layer portion for image classification of the sub-group of the group of categories of images, based on the respective features of the training images and according to classification labels of the training images, wherein the training of the second output layer portion does not comprise training the feature extraction layer portion; and
generating a second DNN model for image classification of the sub-group of the group of categories of images by combining the feature extraction layer portion and the second output layer portion.
15. The non-transitory computer-readable medium according to claim 14 , wherein the first DNN model is training using a subset of the images.
16. The non-transitory computer-readable medium according to claim 14 , wherein the operations further comprise:
receiving a new image; and
classifying, using the second DNN model, the new image.
17. The non-transitory computer-readable medium according to claim 14 , wherein the operations further comprise:
receiving additional training images;
extracting, using the feature extraction layer portion, respective features of the additional training images; and
wherein the training of the second output layer portion comprises training the second output layer portion based further on the respective features of the additional training images.
18. The non-transitory computer-readable medium according to claim 14 , wherein the operations further comprise:
adjusting the feature extraction layer portion and the second output layer portion of the second DNN model based on the training images and the classification labels of the training images.
19. The non-transitory computer-readable medium according to claim 18 , wherein adjusting the feature extraction layer portion and the output layer portion of the second DNN model comprises:
training the second DNN model by using original data of the training images and the classification labels.
20. The non-transitory computer-readable medium according to claim 19 , wherein the adjusting the feature extraction layer portion and the output layer portion of the second DNN model comprises using at least one graphics processing unit.