Method and apparatus for checkout based on image identification technique of convolutional neural network

1. A checkout method based on an image identification technique of convolutional neural network, comprising:

acquiring a plurality of pictures of a single item to be classified placed on a checkout counter, the plurality of pictures corresponding to a plurality of shooting angles one to one;

performing object detection in each of the pictures to obtain a rectangular area image, the rectangular area image being an image corresponding to a rectangular area containing the item;

obtaining a plurality of primary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained first-level classification model, obtaining a first-level classification result of the plurality of pictures according to the plurality of primary classification results and a pre-trained first-level linear regression model, the first-level classification model being a model that is constructed by the image identification technique of convolutional neural network and trained by all the items in a shopping place;

using the first-level classification result as a first classification result; and

performing checkout according to the first classification result.

2. The checkout method according to claim 1 , wherein after the obtaining a plurality of primary classification results and before the using the first-level classification result as a first classification result, the checkout method further comprises:

obtaining a plurality of secondary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained second-level classification model, obtaining a second-level classification result of the plurality of pictures according to the plurality of secondary classification results and a pre-trained second-level linear regression model, and using the second-level classification result as the first classification result, in case that the first-level classification result indicates a similar item; otherwise performing the first-level classification result as a first classification result, the second-level classification model being a model that is constructed by the image identification technique of convolutional neural network and pre-trained by similar items in the shopping place.

3. The checkout method according to claim 1 , wherein after the using the first-level classification result as a first classification result and before the performing checkout according to the first classification result, the checkout method further comprising

determining whether the item in the pictures is consistent with the first classification result, according to a plurality of the rectangular area images and a pre-trained support vector machine model corresponding to the first classification result, in case that the item in the pictures is consistent with the first classification result, performing the checkout step; otherwise, reminding a customer that the item is out of stock.

4. The checkout method according to claim 1 , wherein one camera is disposed in a position just above the item to be classified, so that the item is shot downwards from just above to acquire the picture; and

four cameras are disposed around the item to be classified, so that the item is shot obliquely downwards.

5. A checkout apparatus based on an image identification technique of convolutional neural network, comprising:

an acquisition device for acquiring a plurality of pictures of a single item to be classified placed on a checkout counter, the plurality of pictures corresponding to a plurality of shooting angles one to one;

an object detection device for performing object detection in each of the pictures to obtain a rectangular area image, the rectangular area image being an image corresponding to a rectangular area containing the item;

a first classification device for obtaining a plurality of primary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained first-level classification model, and obtain a first-level classification result of the plurality of pictures according to the plurality of primary classification results and a pre-trained first-level linear regression model, the first-level classification model being a model that is constructed by the image identification technique of convolutional neural network and trained by all the items in the shopping place;

a confirmation device for using the first-level classification result as a first classification result; and

a checkout device for performing checkout according to the first classification result.

6. The checkout apparatus according to claim 5 , further comprising

a second classification device for obtaining a plurality of secondary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained second-level classification model, obtain a second-level classification result of the plurality of pictures according to the plurality of secondary classification results and a pre-trained second-level linear regression model, and use the second-level classification result as the first classification result, in case that the first-level classification result indicates a similar item, the second-level classification model being a model that is constructed by the image identification technique of convolutional neural network and pre-trained by similar items in the shopping place, and

correspondingly, in case that the first-level classification result indicates a similar item, the checkout device performs checkout according to the second-level classification result obtained by the second classification device; otherwise, the checkout device performs checkout according to the first-level classification result obtained by the first classification device.

7. The checkout apparatus according to claim 5 , further comprising

a determination device for determining whether the item in the pictures is consistent with the first classification result according to a plurality of the rectangular area images and a pre-trained support vector machine model corresponding to the first classification result, and

a selection module for calling the checkout device, in case that the item in the pictures is consistent with the first classification result; otherwise, remind a customer that the item is out of stock.

8. The checkout apparatus according to claim 5 , wherein the acquisition device includes one camera disposed in a position just above the item to be classified, so that the item is shot downwards from just above to acquire pictures, and four cameras disposed around the item to be classified, so that the item is shot obliquely downwards.

9. A checkout apparatus based on an image identification technique of convolutional neural network, comprising:

cameras configured to acquire a plurality of pictures of a single item to be classified placed on a checkout counter, the plurality of pictures corresponding to a plurality of shooting angles one to one;

a processor; and

a memory configured to record processor-executable instructions, the processor configured to

perform object detection in each of the pictures to obtain a rectangular area image that is an image corresponding to a rectangular area containing the item,

obtain a plurality of primary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained first-level classification model, obtain a first-level classification result of the plurality of pictures according to the plurality of primary classification results and a pre-trained first-level linear regression model,

use the first-level classification result as a first classification result, and perform checkout according to the first classification result,

the first-level classification model being a model that is constructed by the image identification technique of convolutional neural network and trained by all the items in the shopping place.

10. The checkout apparatus according to claim 9 , wherein the processor is further configured to obtain a plurality of secondary classification results correspondingly, according to a plurality of the rectangular area images and a pre-trained second-level classification model, obtain a second-level classification result of the plurality of pictures according to the plurality of secondary classification results and a pre-trained second-level linear regression model, and use the second-level classification result as the first classification result, in case that the first-level classification result indicates a similar item, the second-level classification model being a model that is constructed by the image identification technique of convolutional neural network and pre-trained by similar items in the shopping place; otherwise, use the first-level classification result as the first classification result.