Animal face recognition system and method based on integrated learning

Abstract

This disclosure provides an integrated learning-based animal face recognition system and method. By analyzing the same image using different recognition methods and employing a majority voting method supplemented by a confidence level, the system identifies pets. This not only eliminates the possibility of misjudgment that may arise from using a single method to identify pets, but also allows for accurate monitoring of the daily routines of each pet in a multi-pet system. Furthermore, by accurately identifying each pet, it enables timely veterinary care when potential health abnormalities occur, thus securing the best medical opportunity and providing pets with a better quality of life.

Claims

1. An integrated learning-based animal face recognition system, comprising: camera; Storage device, storing multiple modules and databases; The system includes a processor, coupled to the camera and the storage, and configured to: acquire multiple images via the camera and store the images in the database; execute a prediction classification module within the modules to apply a first classification to the images to obtain multiple first classification results, and to apply a second classification to the images to obtain multiple second classification results, wherein each of the first classification results corresponds to a first confidence level score, and each of the second classification results corresponds to a second confidence level score; execute an integration voting module within the modules to obtain a first category based on the first classification results, and a second category based on the second classification results, wherein the first category corresponds to a first category confidence level, and the second category corresponds to a second category confidence level; execute a classification decision module within the modules to determine the category corresponding to the larger of the first category confidence level and the second category confidence level as the animal face identity corresponding to the images; and group the images to obtain multiple groups. For each image in each of the groups, the first classification and the second classification are used respectively to obtain a first classification result and a second classification result for each image; a vote is taken on each first classification result in each group to obtain a first classification sub-result, and the first confidence level scores corresponding to each first classification result are averaged to obtain a first confidence sub-score; a vote is taken on each second classification result in each group to obtain a second classification sub-result, and the second confidence level scores corresponding to each second classification result are averaged to obtain a second confidence sub-score; the classification sub-result corresponding to the larger of the first confidence sub-score and the second confidence sub-score is selected as the group result; in response to the fact that each first classification result in the first group of the groups is different and each first confidence level score is the same, the first classification sub-result is set to an empty set; In response to the fact that each of the second classification results of the first group in the groups is different and each of the second confidence level scores is the same, the second classification sub-result is set to the empty set; and in response to the fact that both the first classification sub-result and the second classification sub-result are the empty set, the group result of the first group is set to the empty result.

2. The integrated learning animal face recognition system as claimed in claim 1, wherein the processor is further configured to: execute the animal face detection module in the modules to acquire multiple animal face images from the images, and store the animal face images in the database, wherein each of the animal face images corresponds to a unique identifier.

3. The integrated learning animal face recognition system as described in claim 2, wherein the processor is further configured to: select multiple regions of interest from the images, wherein each of the regions of interest includes a target animal face.

4. The integrated learning animal face recognition system as described in claim 3, wherein the processor is further configured to: obtain the regions of interest through at least one of edge computing, box selection, and circle selection.

5. The integrated learning animal face recognition system as claimed in claim 1, wherein the processor is further configured to: execute the animal face training module among the modules to train the prediction classification module based on the images.

6. The integrated learning animal face recognition system as claimed in claim 1, wherein the processor is further configured to: obtain the first classification results by using Local Binary Patterns (LBP) and K Nearest Neighbor (KNN) as the first classification; and obtain the second classification results by using a Convolutional Neural Network (CNN) as the second classification.

7. The integrated learning animal face recognition system as claimed in claim 1, wherein the processor is further configured to: output an empty group result in response to the first confidence sub-score being the same as the second confidence sub-score and the first sub-class result being different from the second sub-class result.

8. The integrated learning animal face recognition system as claimed in claim 1, wherein the processor is further configured to: in response to obtaining the animal face identity, annotate the images with the animal face identity to obtain a plurality of annotated images, and output the annotated images.

9. An integrated learning-based animal face recognition method, comprising: Multiple images are acquired using a camera and stored in a database. A first classification is applied to these images to obtain multiple first classification results, and a second classification is applied to them to obtain multiple second classification results. Each of the first classification results corresponds to a first confidence level score, and each of the second classification results corresponds to a second confidence level score. A first category is obtained based on the first classification results, and a second category is obtained based on the second classification results, where the first category corresponds to a first category confidence level, and the second category corresponds to a second category confidence level. The category corresponding to the larger of the first category confidence level and the second category confidence level is used as the animal face identity corresponding to these images. The images are grouped to obtain multiple groups. For each image in each group, the first classification and the second classification are applied respectively to obtain a first classification result and a second classification result for each image. A vote is taken on each first classification result in each group to obtain a first sub-result, and the average of each first confidence level score corresponding to each first classification result is obtained to obtain a first confidence sub-score. Voting is performed on each second classification result in each group to obtain a second sub-classification result, and the average of each second confidence level score corresponding to each second classification result is performed to obtain a second confidence sub-score; the sub-classification result corresponding to the larger of the first confidence sub-score and the second confidence sub-score is selected as the group result; in response to the fact that each first classification result in the first group of these groups is different and each first confidence level score is the same, the first sub-classification result is set to an empty set; in response to the fact that each second classification result in the first group of these groups is different and each second confidence level score is the same, the second sub-classification result is set to the empty set; and in response to the fact that both the first sub-classification result and the second sub-classification result are the empty set, the group result of the first group is set to an empty result.

10. The integrated learning-based animal face recognition method as described in claim 9 further includes: Multiple animal face images are obtained from these images and stored in the database, with each animal face image corresponding to a unique identification code.

11. The integrated learning method for animal face recognition as described in claim 10 further includes: Multiple regions of interest were selected from these images, each of which included a target animal's face.

12. The integrated learning-based animal face recognition method as described in claim 11 further includes: The regions of interest are obtained through at least one of the following operations: edge operations, selection of squares, and selection of circles.

13. The integrated learning-based animal face recognition method as described in claim 9 further includes: The first category and the second category are trained based on these images.

14. The integrated learning-based animal face recognition method as described in claim 9 further includes: The first classification results are obtained by using local binary patterns and the K-nearest neighbor algorithm as the first classification. And by using convolutional neural networks as the second classification to obtain these second classification results.

15. The integrated learning-based animal face recognition method as described in claim 9 further includes: In response to the first confidence sub-score being the same as the second confidence sub-score, and the first sub-class result being different from the second sub-class result, the group result is output as an empty result.

16. The integrated learning-based animal face recognition method as described in claim 9 further includes: In response to obtaining the animal's facial identity, the images are labeled with the animal's facial identity to obtain multiple annotated images, and the annotated images are output.

Images