Search

CN-115457528-B - Image recognition method, storage medium, system and equipment

CN115457528BCN 115457528 BCN115457528 BCN 115457528BCN-115457528-B

Abstract

The invention provides an image recognition method, a storage medium, a system and equipment. The method comprises the steps of obtaining an article image, inputting the article image into an image recognition model, obtaining a rectangular recognition frame B p of an article generated by the image recognition model on the article image, obtaining a recognition frame B k to be confirmed from the recognition frame B p , obtaining sub recognition frames B kx from different positions in the recognition frame B k to be confirmed, obtaining a predicted article type T kx of each sub recognition frame B kx , obtaining an article type ZT k with the largest occurrence number in all the predicted article types T kx , and determining the article type output by the recognition frame B k to be confirmed according to the article types ZT k . Wherein the storage medium, system and apparatus are capable of implementing the above method. By the arrangement, the accuracy of the identification result can be improved.

Inventors

  • MA JIAN
  • BI YANHUA
  • KONG LINGLEI

Assignees

  • 青岛海尔电冰箱有限公司
  • 海尔智家股份有限公司

Dates

Publication Date
20260512
Application Date
20220831

Claims (12)

  1. 1. An image recognition method for a refrigeration appliance, comprising: acquiring an article image; Inputting the article image into an image recognition model, and obtaining rectangular recognition frames B p ,p={1,2,…,n 1 },n 1 of the article generated by the image recognition model on the article image as the number of the recognition frames B p ; Acquiring an identification frame B k to be confirmed from the identification frames B p , identifying the identification frames B p to be confirmed, and acquiring the predicted article type T p and the confidence coefficient C p of each identification frame B p ; the recognition frame B p of the confidence level C p between a first threshold and a second threshold is the recognition frame B k to be confirmed; If the confidence coefficient C p is greater than the first threshold, the item type output by the identification frame B p corresponding to the confidence coefficient C p is the predicted item type T p ; If the confidence coefficient C p is smaller than the second threshold value, the identification frame B p corresponding to the confidence coefficient C p does not output the article type; Obtaining sub-identification frames B kx , x={1,2,…,n 3 },n 3 from different positions in the identification frame B k to be confirmed as the number of the sub-identification frames B kx in the identification frame B k to be confirmed; Acquiring a predicted article type T kx of each sub-identification frame B kx ; Acquiring an article type ZT k with the largest occurrence number in all the predicted article types T kx ; acquiring a total number sumT k of occurrences of the item category ZT k in all of the predicted item categories T kx ; And determining the type of the article output by the identification frame B k to be confirmed according to the total times sumT k .
  2. 2. The image recognition method of claim 1, further comprising: If the total number sumT k is smaller than the reference value, the item type output by the identification frame to be confirmed B k is the predicted item type T p of the identification frame B p corresponding to the identification frame to be confirmed B k .
  3. 3. The image recognition method of claim 2, wherein, If the total number sumT k is greater than or equal to the reference value and only one article type ZT k exists, the article type output by the identification to be confirmed box B k is the article type ZT k .
  4. 4. The image recognition method of claim 2, wherein, If the total number sumT k is greater than or equal to a reference value, and there are two or more article types ZT ky ,y={1,2,…,n 4 },2≤ n 4 ＜n 3 ,n 4 with the largest number of occurrences in parallel as the number of article types ZT ky ; acquiring the confidence coefficient C kx and sumC ky of the confidence coefficient C kx of the predicted article type T kx corresponding to each article type ZT ky ; The article type output by the identification frame to be confirmed B k is the maximum confidence and the article type ZT ky corresponding to sumC ky .
  5. 5. The image recognition method of claim 1, wherein, The first threshold is any value between 0.75 and 0.85, and the second threshold is any value between 0.4 and 0.5.
  6. 6. The image recognition method of claim 1, wherein, The number n 3 of the sub-identification frames B kx is any integer between 4 and 8.
  7. 7. The image recognition method of claim 2, wherein, The reference value is a first integer greater than or equal to n 3 /2, the area of the sub-identification frame B kx is 50% -90% of the area of the identification frame B k to be confirmed, and different sub-identification frames B kx can be partially overlapped.
  8. 8. The image recognition method according to claim 1, wherein all of the sub-recognition frames B kx include sub-recognition frames symmetrical with respect to a center line of the recognition frame B k to be confirmed.
  9. 9. The image recognition method according to claim 1, wherein all the sub-recognition frames B kx include sub-recognition frames attached to each side line and each top corner of the recognition frame B k to be confirmed in different directions.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of the image recognition method according to any one of claims 1-9.
  11. 11. An identification system for a refrigeration appliance, the identification system comprising a memory and a processor, the memory storing a computer program executable on the processor, the processor implementing the steps in the image identification method of any one of claims 1-9 when the computer program is executed.
  12. 12. A refrigeration appliance comprising an identification system including a memory and a processor, the memory storing a computer program executable on the processor, the processor implementing the steps in the image identification method of any one of claims 1 to 9 when the computer program is executed.

Description

Image recognition method, storage medium, system and equipment Technical Field The present invention relates to the field of household appliances, and in particular, to an image recognition method, a storage medium, a system, and an apparatus for a refrigeration apparatus. Background Along with the progress of science and technology, the requirements of users on refrigeration equipment are higher and higher, and intelligent transformation becomes a new research and development direction of the refrigeration equipment. The conventional intelligent refrigerating equipment is generally provided with an identification system, and when an image of an article stored in the refrigerating equipment is acquired, an identification frame can be generated in the image, and the identification frame forms a candidate area to identify the article. In the prior art, the candidate area formed by the identification frame is often directly generated into the predicted article type corresponding to the identification frame. However, this design has the disadvantage of being prone to false identifications. Disclosure of Invention The invention aims to provide an image recognition method, a storage medium, a system and equipment, which can improve the accuracy of recognition results by acquiring sub-recognition frames B kx from different positions in a recognition frame B k to be confirmed and acquiring the predicted article type T kx of each sub-recognition frame B kx. In order to achieve the above object, an embodiment of the present invention provides an image recognition method, including: acquiring an article image; Inputting the article image into an image recognition model, and obtaining rectangular recognition frames B p,p={1,2,…,n1},n1 of the article generated by the image recognition model on the article image as the number of the recognition frames B p; Acquiring a recognition frame B k to be confirmed from the recognition frame B p, wherein the recognition frame k epsilon p to be confirmed; obtaining sub-identification frames B kx, x={1,2,…,n3},n3 from different positions in the identification frame B k to be confirmed as the number of the sub-identification frames B kx in the identification frame B k to be confirmed; Acquiring a predicted article type T kx of each sub-identification frame B kx; Acquiring an article type ZT k with the largest occurrence number in all the predicted article types T kx; And determining the type of the article output by the identification frame B k to be confirmed according to the type ZT k of the article. As a further improvement of an embodiment of the present invention, the "obtaining the identification frame B k to be confirmed from the identification frame B p" includes: identifying the identification frames B p, and acquiring the predicted article type T p and the confidence coefficient C p of each identification frame B p; The identification box B p with the confidence level C p between the first threshold and the second threshold is the identification box B k to be confirmed. As a further improvement of an embodiment of the present invention, the method further comprises: If the confidence coefficient C p is greater than the first threshold, the item type output by the identification frame B p corresponding to the confidence coefficient C p is the predicted item type T p; If the confidence level C p is smaller than the second threshold value, the identification frame B p corresponding to the confidence level C p does not output the article type. As a further improvement of an embodiment of the present invention, the method further comprises: acquiring a total number sumT k of occurrences of the item category ZT k in all of the predicted item categories T kx; If the total number sumT k is smaller than the reference value, the item type output by the identification frame to be confirmed B k is the predicted item type T p of the identification frame B p corresponding to the identification frame to be confirmed B k. As a further improvement of an embodiment of the present invention, wherein, If the total number sumT k is greater than or equal to the reference value and only one article type ZT k exists, the article type output by the identification to be confirmed box B k is the article type ZT k. As a further improvement of an embodiment of the present invention, wherein, If the total number sumT k is greater than or equal to a reference value, and there are two or more article types ZT ky,y={1,2,…,n4},2≤ n4＜n3,n4 with the largest number of occurrences in parallel as the number of article types ZT ky; acquiring the confidence coefficient C kx and sumC ky of the confidence coefficient C kx of the predicted article type T kx corresponding to each article type ZT ky; The article type output by the identification frame to be confirmed B k is the maximum confidence and the article type ZT ky corresponding to sumC ky. As a further improvement of an embodiment of the present invention, whe