Image recognition method, storage medium, system and device

Abstract

This invention provides an image recognition method, storage medium, system, and device. The method includes: acquiring an image of an object; inputting the image of the object into an image recognition model, and acquiring a rectangular bounding box B of the object generated by the image recognition model based on the image of the object. p From the identification box B p Obtain the recognition box B to be confirmed k The identification box to be confirmed is k∈p; from the identification box to be confirmed B k Obtain sub-recognition boxes B at different locations within the box. kx ; Obtain each of the sub-identification boxes B kx Predicted item type T kx ; Obtain all predicted item types T kx The most frequently appearing item type (ZT) k According to the item type ZT k Determine the identification box B to be confirmed k The types of items output. The storage medium, system, and device must be capable of implementing the above method. This configuration improves the accuracy of the recognition results.

Description

Technical Field

[0001] This invention relates to the field of household appliances, and more particularly to an image recognition method, storage medium, system, and device for refrigeration equipment. Background Technology

[0002] With the advancement of technology, users have increasingly higher requirements for refrigeration equipment, making intelligent transformation a new research and development direction for refrigeration equipment. Existing intelligent refrigeration equipment is generally equipped with a recognition system. After acquiring an image of an item stored inside the refrigeration equipment, it can generate a bounding box in the image, and then use the bounding box to form candidate regions for item identification. Current technologies often directly generate the predicted item type corresponding to the bounding box as the candidate region. However, this design has the following drawback: it is prone to misidentification. Summary of the Invention

[0003] The purpose of this invention is to provide an image recognition method, storage medium, system, and device, which, by analyzing the image from the image in the frame B to be identified... k Obtain sub-recognition boxes B at different locations within the box. kx And obtain each of the sub-identification boxes B. kx Predicted item type T kx This can improve the accuracy of the recognition results.

[0004] To achieve the above-mentioned objective, one embodiment of the present invention provides an image recognition method, comprising:

[0005] Get the item image;

[0006] The image of the item is input into an image recognition model to obtain a rectangular bounding box B of the item generated by the image recognition model. p p={1,2,…,n1}, where n1 is the recognition box B p The number of;

[0007] From the recognition box B p Obtain the recognition box B to be confirmed k The bounding box to be confirmed is k∈p;

[0008] From the identification box B to be confirmed k Obtain sub-recognition boxes B at different locations within the box. kx x={1,2,…,n3}, where n3 is the bounding box B to be confirmed. k The sub-identification box B within the frame kx The number of;

[0009] Obtain each of the sub-identification boxes B kx Predicted item type T kx ;

[0010] Obtain all predicted item types T kx The most frequently appearing item type (ZT) k ;

[0011] According to the item type ZT k Determine the identification box B to be confirmed k The types of items output.

[0012] As a further improvement to one embodiment of the present invention, wherein, "from the identification box B" p Obtain the recognition box B to be confirmed k "include:

[0013] Identify the identification box B p And obtain each of the aforementioned recognition boxes B p Predicted item type T p and confidence level C p ;

[0014] The confidence level C p The recognition box B between the first threshold and the second threshold p For the identification box B to be confirmed k .

[0015] As a further improvement to one embodiment of the present invention, it further includes:

[0016] If the confidence level C p If the confidence level C is greater than the first threshold, then the confidence level C p The corresponding recognition box B p The output item type is the predicted item type T. p ;

[0017] If the confidence level C p If the confidence level C is less than the second threshold, then the confidence level C p The corresponding recognition box B p The item type is not output.

[0018] As a further improvement to one embodiment of the present invention, it further includes:

[0019] Obtain all predicted item types T kx The types of items described in ZT k The total number of occurrences, sumT k ;

[0020] If the total number of times sumT k If it is less than the baseline value, then the identification box B to be confirmed k The output item type is the identification box B to be confirmed. k The corresponding recognition box Bp The predicted item type T p .

[0021] As a further improvement to one embodiment of the present invention, wherein,

[0022] If the total number of times sumT k Greater than or equal to the baseline value, and there exists only one of the item types ZT. k Then the identification box B to be confirmed k The output item type is the item type ZT. k .

[0023] As a further improvement to one embodiment of the present invention, wherein,

[0024] If the total number of times sumT k The item type ZT that is greater than or equal to the baseline value and has two or more items with the same number of occurrences. ky y = {1, 2, ..., n4}, 2 ≤ n4 < n3, where n4 is the type of item ZT. ky The number of;

[0025] Get each of the aforementioned item types ZT ky The corresponding predicted item type T kx confidence level C kx confidence level and sumC ky ;

[0026] The identification box to be confirmed B k The output item types are the maximum stated confidence level and sumC. ky The corresponding item type ZT ky .

[0027] As a further improvement to one embodiment of the present invention, wherein,

[0028] 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.

[0029] As a further improvement to one embodiment of the present invention, wherein,

[0030] The sub-identification box B kx The number of elements n3 is any integer between 4 and 8.

[0031] As a further improvement to one embodiment of the present invention, wherein,

[0032] The reference value is the first integer greater than or equal to n³ / 2, and the sub-identification box B kx The area is the identification box B to be confirmed. kThe area varies from 50% to 90%, depending on the specific sub-identification box B. kx They can partially overlap.

[0033] As a further improvement to one embodiment of the present invention, wherein all of the sub-identification boxes B kx Includes the identification box B to be confirmed k Sub-identification boxes symmetrical about the center line.

[0034] As a further improvement to one embodiment of the present invention, wherein all of the sub-identification boxes B kx Includes the identification frame B that is to be confirmed. k Sub-identification boxes for each edge line and each vertex in different directions.

[0035] To achieve the above-mentioned objectives, one embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the image recognition method described in any of the above embodiments.

[0036] To achieve the above-mentioned objectives, one embodiment of the present invention provides an identification system for refrigeration equipment, wherein the identification system includes a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the computer program, it implements the steps of the image recognition method described in any of the above embodiments.

[0037] To achieve the above-mentioned objectives, one embodiment of the present invention provides a refrigeration device, wherein the refrigeration device includes an identification system, the identification system includes a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the computer program, it implements the steps in the image recognition method described in any of the above embodiments.

[0038] Compared with the prior art, the present invention improves upon the identification of the unconfirmed frame B by... k Obtain sub-recognition boxes B at different locations within the box. kx And obtain each of the sub-identification boxes B. kx Predicted item type T kx Its beneficial effect is that it can improve the accuracy of the recognition results. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the structure of a refrigeration device according to an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of an article image and its recognition frame according to an embodiment of the present invention;

[0041] Figure 3This is a schematic diagram of the identification frame to be confirmed and its sub-identification frames of an item image according to an embodiment of the present invention;

[0042] Figure 4 yes Figure 3 The diagram shows the identification box to be confirmed and its sub-identification boxes;

[0043] Figure 5 yes Figure 3 The diagram shows the identification box to be confirmed and its sub-identification boxes;

[0044] Figure 6 This is a flowchart of an image recognition method according to an embodiment of the present invention;

[0045] Figure 7 This is a flowchart of another embodiment of the image recognition method of the present invention. Detailed Implementation

[0046] The present invention will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0047] Reference Figure 1 and Figure 2 In one embodiment of the present invention, the refrigeration equipment 100, such as a refrigerator, freezer, commercial display cabinet, etc., may include a cabinet body 1 and a door body 2. The cabinet body 1 may have a storage compartment for storing items 4 inside, and the door body 2 may be used to open and close the storage compartment. The door body 2 may also be provided with a bottle holder structure for storing items 4, or the door body 2 may also have a door compartment for storing items 4 inside, etc.

[0048] The refrigeration equipment 100 may be equipped with an image recognition module 3, which may include an image recognition model and an image acquisition module. The image acquisition module may include a camera, which can be used to acquire images 5 of the items 4 stored in the refrigeration equipment 100. The image recognition model can be used to perform target detection and category identification on the item images 5. The refrigeration equipment 100 can obtain information such as the category of the items stored in the refrigeration equipment 100 by knowing the detection results output by the image recognition model, so as to manage the items 4 stored in the refrigeration equipment 100.

[0049] Reference Figure 2 In the process of processing the object image 5, the image recognition model can first perform target detection on the object image 5 and generate a rectangular recognition box 6 covering the maximum outline of the target object 4 in the image (refer to...). Figure 2(The rectangle is marked with a dashed line). Target item 4 is the item that the recognition box 6 wants to identify. The image recognition model can be an algorithmic model built using deep convolutional neural networks, etc.

[0050] The recognition box 6 is a box calculated and output by the image recognition model. The location of the recognition box 6 can be identified by marking the upper left and lower right coordinates of its rectangle, i.e., (x... 左上 y 左上 ) and (x 右下 y 右下 Alternatively, its location can be identified by marking the center coordinates of its rectangle and the rectangle's width and height, i.e., (x... 中心 y 中心 w 宽 and h 高 .

[0051] After generating the recognition box 6, the image recognition model can obtain the predicted item type and confidence level corresponding to each recognition box 6 by sending the region formed by each recognition box 6 into the classification and recognition network, or by extracting features from the region formed by each recognition box 6 and comparing them with pre-stored feature data.

[0052] The predicted item type can refer to the name information representing a specific item type predicted by the image recognition model, such as milk, yogurt, etc.

[0053] The confidence level can be a numerical value representing the accuracy of the prediction of the item type. The confidence level can be a value between 0 and 1.

[0054] Reference Figures 2 to 6 In one embodiment of the present invention, an image recognition method is provided, comprising:

[0055] Get item image 5;

[0056] The image of the item 5 is input into the image recognition model to obtain the rectangular recognition box 6B of the item 4 generated by the image recognition model for the image of the item 5. p p={1,2,…,n1}, where n1 is the recognition box 6B p The number of;

[0057] From the recognition frame 6B p Obtain the unconfirmed recognition box 6B from the middle k The bounding box to be confirmed is 6k∈p;

[0058] From the identification frame 6B to be confirmed k Obtain sub-recognition boxes 7B at different locations within the box. kx x={1,2,…,n3}, where n3 is the identification box 6B to be confirmed. kThe sub-identification box 7B within the frame kx The number of;

[0059] Obtain each of the aforementioned sub-identification boxes 7B kx Predicted item type T kx ;

[0060] Obtain all predicted item types T kx The most frequently appearing item type (ZT) k ;

[0061] According to the item type ZT k Determine the identification box 6B to be confirmed. k The types of items output.

[0062] In this embodiment, the sub-identification box 7 can refer to a rectangular box located within the identification box 6 to be confirmed. The area of ​​the sub-identification box 7 is smaller than that of the corresponding identification box 6 to be confirmed. The same identification box 6 to be confirmed may contain a number of sub-identification boxes 7. The sub-identification boxes 7 within the same identification box 6 are different, but they can partially overlap.

[0063] In actual recognition, when the information of the item image 5 contained within the recognition box 6 to be confirmed is diverse, for example... Figure 3 The bounding box 6B3 contains both milk and yogurt image information. If the entire area selected by the bounding box 6, which contains information about multiple item images 5, is identified, the confidence level of the predicted item type is relatively low, and the predicted item type may also be incorrect.

[0064] By selecting several sub-identification boxes 7 at different positions within the identification box 6 to be confirmed, for example... Figures 3 to 5 Sub-identification box 7B corresponding to the identification box 6B3 to be confirmed 31 -B 36 Since the information of the item image 5 covered by the sub-identification box 7 is less and more singular than that of the identification box 6B3 to be confirmed, the confidence level of the predicted item type is relatively high when the area selected by the sub-identification box 7 is identified, and the predicted item type is relatively accurate.

[0065] Since different sub-identification boxes 7 cover different areas, the corresponding predicted item types may be the same or different. Since identification boxes 6 are generated for target item 4, such as yogurt in the identification box to be confirmed 6B3, the identification box to be confirmed 6 generally contains the most image information of target item 4. Therefore, the item type that appears most frequently among the predicted item types of all sub-identification boxes 7 is generally the item type corresponding to the target item 4 that the identification box to be confirmed 6 wants to identify.

[0066] This setup enables accurate identification of target item 4 in item image 5, makes the item type output by the image recognition model more accurate, improves the accuracy of the recognition results, and avoids missed identification and incorrect identification.

[0067] Reference Figures 3 to 5 Furthermore, in another embodiment of the present invention, the image recognition method wherein the sub-recognition box 7B kx The number n3 is any integer between 4 and 8. For example, the sub-identification box 7 corresponding to the identification box 6B3 to be confirmed has B. 31 -B 36 There are 6 in total.

[0068] This setting ensures that an appropriate number of sub-recognition boxes 7 are selected, avoiding incomplete image information coverage due to too few selected sub-recognition boxes 7, which would lead to inaccurate recognition results of the recognition box 6 to be confirmed. At the same time, it avoids excessive computation and long computation time due to too many selected sub-recognition boxes 7, which would result in low recognition efficiency of the image recognition model.

[0069] Reference Figures 3 to 5 Furthermore, in another embodiment of the present invention, the image recognition method wherein the sub-recognition box 7B kx The area is the identification box to be confirmed, 6B. k The area varies from 50% to 90%, depending on the sub-identification box 7B. kx They can partially overlap.

[0070] This setting ensures that the sub-identification box 7 covers an appropriate area, preventing situations where the area covered by the sub-identification box 7 is too small, resulting in insufficient information on the object image 5 contained in the sub-identification box 7 or incomplete image information covered by the set of sub-identification boxes 7. This would lead to inaccurate identification results for both the sub-identification box 7 and the identification box to be confirmed 6. At the same time, it avoids situations where the area covered by the selected sub-identification box 7 is too large, resulting in the information on the object image 5 contained in the sub-identification box 7 being essentially the same as that in the identification box to be confirmed 6, thus reducing the value of identifying the sub-identification box 7.

[0071] Reference Figure 3 Furthermore, in another embodiment of the present invention, the image recognition method wherein all of the sub-recognition boxes 7B kx Includes the identification box 6B to be confirmed k Center line 8 symmetrical sub-identification boxes 7.

[0072] In this embodiment, the sub-identification boxes 7 can be selected in pairs. For example, sub-identification boxes 7 that are symmetrical about the horizontal center line 8 can be selected, or sub-identification boxes 7 that are symmetrical about the vertical center line 8 can be selected.

[0073] This setting allows the selected sub-recognition boxes 7 to be distributed more evenly and reasonably, preventing situations such as multiple sub-recognition areas clustering around a specific area within the recognition box 6 to be confirmed, or the set of sub-recognition boxes 7 covering incomplete image information, thereby ensuring the accuracy of the recognition result of the recognition box 6 to be confirmed.

[0074] Reference Figure 4 and Figure 5 Furthermore, in another embodiment of the present invention, the image recognition method wherein all of the sub-recognition boxes 7B kx Includes the identification frame 6B that is fitted to the object to be confirmed. k Sub-identification boxes 7 for each edge line and each vertex 9 in different directions.

[0075] For example, you can select the sub-recognition boxes 7 that fit the top left, top right, bottom left, and bottom right corners 9 of the recognition box 6 to be confirmed, or you can select the sub-recognition boxes 7 that fit the top, bottom, left, and right sides of the recognition box 6.

[0076] This setting allows the selected sub-recognition boxes 7 to be distributed more evenly and reasonably, preventing situations such as multiple sub-recognition areas clustering around a specific area within the recognition box 6 to be confirmed, or the set of sub-recognition boxes 7 covering incomplete image information, thereby ensuring the accuracy of the recognition result of the recognition box 6 to be confirmed.

[0077] Furthermore, in another embodiment of the present invention, all of the said sub-identification frames 7B kx The frame may contain the identification box 6B to be confirmed. k The central sub-identification box 7.

[0078] Reference Figure 2 and Figure 7 Furthermore, in another embodiment of the present invention, the image recognition method, wherein, "from the recognition frame 6B" p Obtain the unconfirmed recognition box 6B from the middle k "include:

[0079] Identify the identification frame 6B p And obtain each of the aforementioned recognition boxes 6B p Predicted item type T p and confidence level C p ;

[0080] The confidence level C p The recognition box 6B between the first threshold and the second threshold p The identification box to be confirmed is 6B. k .

[0081] In this embodiment, each recognition box 6 generated by the image recognition model can be recognized as a whole first to obtain the predicted item type and confidence level of each recognition box 6. Based on the confidence level, the recognition boxes 6 to be confirmed are initially selected. For the recognition boxes 6 to be confirmed, the steps such as selecting sub-recognition areas in the above embodiment are performed to confirm the item type output by the image recognition model.

[0082] This setup reduces computational load and time, improves the efficiency of image recognition models, and ensures the accuracy and reliability of recognition results, avoiding misidentification and missed identification.

[0083] Reference Figure 2 and Figure 7 Furthermore, in another embodiment of the present invention, the image recognition method further includes:

[0084] If the confidence level C p If the confidence level C is greater than the first threshold, then the confidence level C p The corresponding recognition frame 6B p The output item type is the predicted item type T. p ;

[0085] If the confidence level C p If the confidence level C is less than the second threshold, then the confidence level C p The corresponding recognition frame 6B p The item type is not output.

[0086] In this embodiment, if the confidence level C p If the confidence level is greater than the first threshold, it indicates that the confidence level C is greater than the first threshold. p Corresponding recognition box 6B p The predicted item type is highly likely to be the same as the item type corresponding to target item 4, and therefore can be used as the recognition box 6B. p The types of items output; if the confidence level is C p If the confidence level is less than the second threshold, it indicates that the confidence level C is... p Corresponding recognition box 6B p It might be incorrect; the recognition box is 6B. p Item 4 may not have been correctly selected. For example, frame 6 might have selected the container holding item 4, so there's no need to perform further recognition on frame 6. Therefore, frame 6B... p The item type is not output.

[0087] This setup reduces computational load and time, improves the efficiency of image recognition models, and ensures the accuracy and reliability of recognition results, avoiding misidentification and missed identification.

[0088] Furthermore, in another embodiment of the present invention, the image recognition method 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.

[0089] Reference Figure 7 Furthermore, in another embodiment of the present invention, the image recognition method further includes:

[0090] Obtain all predicted item types T kx The types of items described in ZT k The total number of occurrences, sumT k ;

[0091] If the total number of times sumT k If it is less than the benchmark value, then the identification box 6B to be confirmed k The output item type is the identification box 6B to be confirmed. k The corresponding recognition frame 6B p The predicted item type T p .

[0092] In actual identification, there may be situations where a certain item category appears most frequently among all predicted item categories, but its total frequency is relatively low. For example, if there are 8 predicted item categories, only 2 of them may be the same, while the rest may be different.

[0093] Because when the total number of predictions is small, the image information covered by the corresponding sub-identification box 7 is limited, and the total number of predicted item types T is limited. kx Z, the item type that appears most frequently Tk It is very likely not the recognition box 6B to be confirmed. k The target item 4 corresponds to the item type, and therefore can be represented by the identification box 6B to be confirmed. k Global information, pending confirmation recognition box 6B k The corresponding recognition frame 6B p The predicted item type T p The type of item output.

[0094] This setting ensures the accuracy and reliability of the image recognition model's recognition results, avoiding misidentification and missed identification.

[0095] Furthermore, in another embodiment of the present invention, the image recognition method wherein the reference value is a first integer greater than or equal to n³ / 2.

[0096] For example, when sub-identification box 7B kxWhen the number of elements n3 is 6, the reference value can be 3; when the sub-identification box 7B kx When the number of elements n3 is 7, the reference value can be 4.

[0097] This setting ensures the accuracy and reliability of the image recognition model's recognition results, avoiding misidentification and missed identification.

[0098] Reference Figure 7 Furthermore, in another embodiment of the present invention, the image recognition method, wherein,

[0099] If the total number of times sumT k Greater than or equal to the baseline value, and there exists only one of the item types ZT. k Then the identification box 6B to be confirmed k The output item type is the item type ZT. k .

[0100] This setting ensures the accuracy and reliability of the image recognition model's recognition results, avoiding misidentification and missed identification.

[0101] Reference Figures 3 to 5 , Figure 7 Furthermore, in another embodiment of the present invention, the image recognition method, wherein,

[0102] If the total number of times sumT k The item type ZT that is greater than or equal to the baseline value and has two or more items with the same number of occurrences. ky y = {1, 2, ..., n4}, 2 ≤ n4 < n3, where n4 is the type of item ZT. ky The number of;

[0103] Get each of the aforementioned item types ZT ky The corresponding predicted item type T kx confidence level C kx confidence level and sumC ky ;

[0104] The identification frame to be confirmed 6B k The output item types are the maximum stated confidence level and sumC. ky The corresponding item type ZT ky .

[0105] For example Figures 3 to 5 In the middle, predict the type of item T 31 - T 36 In the data, milk and yogurt both appeared 3 times, tying for the most occurrences, but their confidence levels were significantly different. This was to ultimately determine the identification box 6B to be confirmed. kThe confidence scores for each type of item output can be calculated separately. For example, ZT 31 It's milk, ZT 32 It's yogurt.

[0106] sumC 31 =C 32 +C 33 +C 35 =0.54+0.6+0.7=1.84.

[0107] sumC 32 =C 31 +C 34 +C 36 =0.6+0.8+0.75=2.15.

[0108] Due to sumC 32 Greater than sumC 31 Therefore, the item type output by the recognition box 6B3 is yogurt.

[0109] This setup ensures the accuracy and reliability of the image recognition model's results, avoids misidentification and omissions, and is simple and convenient.

[0110] In one embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the image recognition method described in any of the above embodiments.

[0111] In one embodiment of the present invention, an identification system for a refrigeration device 100 is provided, wherein the identification system includes a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the computer program, it implements the steps of the image recognition method described in any of the above embodiments.

[0112] In one embodiment of the present invention, a cooling device 100 is provided, wherein the cooling device 100 includes an identification system, the identification system includes a memory and a processor, the memory stores a computer program that can run on the processor, and when the processor executes the computer program, it implements the steps in the image recognition method described in any of the above embodiments.

[0113] In summary, the image recognition method, storage medium, system, and device of the present invention can recognize images from the image frame B to be confirmed. k Obtain sub-recognition boxes B at different locations within the box. kx And obtain each of the sub-identification boxes B. kx Predicted item type T kxThis addresses the problem in existing technologies where predicted item types with confidence levels greater than a set threshold are selected as the item types for the candidate box output, while predicted item types with confidence levels less than the set threshold are not output. When the threshold is set too high, this can easily lead to missed identifications, and when the threshold is set too low, it can easily lead to incorrect identifications.

[0114] The technical solution in this application can improve the accuracy and reliability of the recognition results, avoid misidentification and omission, and at the same time reduce the amount of computation, reduce the computation time, and improve the recognition efficiency of the image recognition model.

[0115] It should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

[0116] The detailed descriptions listed above are merely specific descriptions of feasible embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. An image recognition method for refrigeration equipment, characterized in that, include: Get the item image; The image of the item is input into an image recognition model to obtain a rectangular bounding box B of the item generated by the image recognition model. p p={1,2,…,n1}, where n1 is the recognition box B p The number of; From the recognition box B p Obtain the recognition box B to be confirmed k The recognition box to be confirmed is k∈p, and the recognition box B is identified. p And obtain each of the aforementioned recognition boxes B p Predicted item type T p and confidence level C p ; The confidence level C p The recognition box B between the first threshold and the second threshold p For the identification box B to be confirmed k ; If the confidence level C p If the confidence level C is greater than the first threshold, then the confidence level C p The corresponding recognition box B p The output item type is the predicted item type T. p ; If the confidence level C p If the confidence level C is less than the second threshold, then the confidence level C p The corresponding recognition box B p Do not output item type; From the identification box B to be confirmed k Obtain sub-recognition boxes B at different locations within the box. kx x={1,2,…,n3}, where n3 is the bounding box B to be confirmed. k The sub-identification box B within the frame kx The number of; Obtain each of the sub-identification boxes B kx Predicted item type T kx ; Obtain all predicted item types T kx The most frequently appearing item type (ZT) k ; Obtain all predicted item types T kx The types of items described in ZT k The total number of occurrences, sumT k ; Based on the total number sumT k Determine the identification box B to be confirmed k The types of items output.

2. The image recognition method as described in claim 1, characterized in that, Also includes: If the total number of times sumT k If it is less than the baseline value, then the identification box B to be confirmed k The output item type is the identification box B to be confirmed. k The corresponding recognition box B p The predicted item type T p .

3. The image recognition method as described in claim 2, characterized in that, If the total number of times sumT k Greater than or equal to the baseline value, and there exists only one of the item types ZT. k Then the identification box B to be confirmed k The output item type is the item type ZT. k .

4. The image recognition method as described in claim 2, characterized in that, If the total number of times sumT k ZT Items that are greater than or equal to the baseline value and have two or more items that have the same number of occurrences as the most frequent. ky y = {1, 2, ..., n4}, 2 ≤ n4 < n3, where n4 is the type of item ZT. ky The number of; Get each of the aforementioned item types ZT ky The corresponding predicted item type T kx confidence level C kx confidence level and sumC ky ; The identification box to be confirmed B k The output item types are the maximum stated confidence level and sumC. ky The corresponding item type ZT ky .

5. The image recognition method as described in claim 1, characterized in that, 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. The image recognition method as described in claim 1, characterized in that, The sub-identification box B kx The number of elements n3 is any integer between 4 and 8.

7. The image recognition method as described in claim 2, characterized in that, The reference value is the first integer greater than or equal to n³ / 2, and the sub-identification box B kx The area is the identification box B to be confirmed. k The area varies from 50% to 90%, depending on the specific sub-identification box B. kx They can partially overlap.

8. The image recognition method as described in claim 1, characterized in that, All of the sub-identification boxes B kx Includes the identification box B to be confirmed k Sub-identification boxes symmetrical about the center line.

9. The image recognition method as described in claim 1, characterized in that, All of the sub-identification boxes B kx Includes the identification frame B that is to be confirmed. k Sub-identification boxes for each edge line and each vertex in different directions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the image recognition method according to any one of claims 1-9.

11. An identification system for refrigeration equipment, characterized in that, The recognition system includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the image recognition method according to any one of claims 1-9.

12. A refrigeration device, characterized in that, The cooling device includes an identification system, which includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the image recognition method according to any one of claims 1-9.

Images