Image Recognition Method, Storage Medium, System and Device for Strip-shaped Articles

Abstract

The present invention provides an image recognition method, a storage medium, a system and a device for strip-shaped articles. The method includes: obtaining an article image containing strip-shaped articles; inputting the article image into an image recognition model to obtain a strip-shaped article recognition area of the strip-shaped articles; obtaining a first sub-recognition area and a second sub-recognition area that are in a long strip shape from the strip-shaped article recognition area, the length direction of the first sub-recognition area extends along a first diagonal line, and the length direction of the second sub-recognition area extends along a second diagonal line; obtaining the predicted article type and confidence level of the first sub-recognition area, and obtaining the predicted article type and confidence level of the second sub-recognition area; obtaining the maximum confidence level, and if the maximum confidence level is greater than a first threshold, the article type output by the strip-shaped article recognition area is the predicted article type corresponding to the maximum confidence level. With such a setting, accurate recognition of strip-shaped articles can be achieved.

Description

Technical Field

[0001] The present invention relates to the field of household appliances, and particularly to an image recognition method, a storage medium, a system and a device for strip-shaped objects. Background Art

[0002] With the progress of technology, users' requirements for refrigeration equipment are getting higher and higher, and the intelligent transformation has become a new research and development direction for refrigeration equipment. Intelligent refrigeration equipment is generally equipped with an image recognition system. The image recognition system usually includes an image acquisition module and an image recognition module. The image acquisition module is used to acquire the images of the objects stored in the refrigeration equipment, and the image recognition module is used to identify the objects in the object images. Through the image recognition system, the refrigeration equipment can know and manage the objects stored in its accommodation space. In the prior art, the image acquisition module generally includes an image recognition model. After the object image is input into the image recognition model, the image recognition model can generate rectangular recognition areas around each object in the object image, and the objects in the object image are recognized through the recognition areas. In the prior art, the recognition areas are often directly recognized and the predicted object types are output. However, this design has the following defects: for strip-shaped objects, due to the large difference in the length and width dimensions, the recognition areas often contain the images of other objects, and the direct recognition method is prone to misidentify the strip-shaped objects. Summary of the Invention

[0003] The purpose of the present invention is to provide an image recognition method, a storage medium, a system and a device for strip-shaped objects, which can accurately identify strip-shaped objects by respectively obtaining a strip-shaped first sub-recognition area and a second sub-recognition area at the two diagonals in the strip-shaped object recognition area.

[0004] To achieve the above invention purpose, an embodiment of the present invention provides an image recognition method for strip-shaped objects, which includes:

[0005] Obtain an object image containing a strip-shaped object;

[0006] Input the object image into an image recognition model to obtain a strip-shaped object recognition area b k , k = {1, 2,..., n1}, n1 is the number of the strip-shaped object recognition areas, and the strip-shaped object recognition area b k is a rectangular area generated by the image recognition model at the maximum contour line of the object. The rectangular strip-shaped object recognition area includes an intersecting first diagonal and a second diagonal;

[0007] Obtain a long strip-shaped first sub-recognition area b k and a second sub-recognition area b k1u and the second sub-recognition area b k2w, u = {1, 2, …, n2}, where n2 is the number of the first sub - recognition area b k1u , w = {1, 2, …, n3}, where n3 is the number of the second sub - recognition area b k2w .

[0008] The length direction of the first sub - recognition area b k1u extends along the first diagonal, and at least part of the first diagonal is located within the first sub - recognition area b k1u .

[0009] The length direction of the second sub - recognition area b k2w extends along the second diagonal, and at least part of the second diagonal is located within the second sub - recognition area b k2w ;

[0010] Obtain the predicted variety type T k1u and confidence level C k1u of the first sub - recognition area b k1u , and obtain the predicted variety type T k2w and confidence level C k2w of the second sub - recognition area b k2w ;

[0011] Obtain the maximum confidence level ZC k1u from the confidence level C k2w . If the maximum confidence level ZC k is greater than the first threshold, the variety type output by the strip - shaped object recognition area b k is the predicted variety type corresponding to the maximum confidence level ZC k . k

[0012] As a further improvement of an embodiment of the present invention, among them, the four vertices of the rectangular strip - shaped object recognition area b k are A k , B k , C k , D k , the first diagonal is A k C k , the second diagonal is B k D k , the boundary line A k C k between the side of the first diagonal A k B k and the boundary line B k C k include points M k1 , point M k2 , point N k1 , point N k2 ​, the point M k1 and the point M k2 are located on the boundary line A k B k . The point N k1 and the point N k2 are located on the boundary line B k C k . The distance from the point M k1 to the vertex A k is L(A k M k ). The distance from the point M k2 to the vertex A k2 is L(A k M k2 ). The length of the boundary line A k B k is L(A k B k ). L(A k M k1 ) = 0.1 * L(A k B k ). L(A k M k2 ) = 0.5 * L(A k B k ). The distance from the point N k1 to the vertex C k is L(C k N k1 ). The distance from the point N k2 to the vertex C k is L(C k N k2 ). The length of the boundary line B k C k is L(B k C k ). L(C k N k1 ) = 0.1 * L(B k C k ). L(C k N k2 ) = 0.5 * L(B k C k ). The point M k1 and the point N k1 form a boundary line M k1 N k1 . The point M k2 and the point N k2 form a boundary line M k2 N k2 . The first sub-identification area b k1uTowards the boundary line A k B k and the boundary line B k C k The boundary lines of are located on the boundary line M k1 N k1 and the boundary line M k2 N k2 between them.

[0013] As a further improvement of an embodiment of the present invention, wherein, the boundary line A k C k on the other side of, and the boundary line D k D k include point P k C k include point P k1 and point P k2 and point Q k1 and point Q k2 wherein, the point P k1 and the point P k2 are located on the boundary line A k D k the point Q k1 and the point Q k2 are located on the boundary line D k C k the distance from the point P k1 to the vertex A k [[ID=##]]is L(A k P k1 ), the distance from the point P k2 to the vertex A k is L(A k P k2 ), the length of the boundary line A k D k is L(A k D k ), L(A k P k1 ) = 0.1 * L(A k D k ), L(A k P k2 ) = 0.5 * L(A k D k ), the distance from the point Q k1 to the vertex C k is L(C k Qy k1 ), the distance from the point Q k2 to the vertex C k is L(C k Q k2 ), the boundary line Dk C k with a length of L(D k C k ), L(C k Q k1 ) = 0.1 * L(D k C k ),L(C k Q k2 ) = 0.5 * L(D k C k ), the point P k1 and the point Q k1 form a boundary line P k1 Q k1 ,the point P k2 and the point Q k2 form a boundary line P k2 Q k2 ,the first sub - recognition area b k1u faces the boundary line A k D k and the boundary line D k C k whose boundary line is located between the boundary line P k1 Q k1 and the boundary line P k2 Q k2 .

[0014] As a further improvement of an embodiment of the present invention, where the boundary line A k D k on one side of the second diagonal B k B k and the boundary line A k D k include the point E k1 、the point E k2 、the point H k1 、the point H k2 ,the point E k1 and the point E k2 are located on the boundary line A k B k ,the point H k1 and the point H k2 are located on the boundary line A k D k ,the distance from the point E k1 to the vertex B k is L(B k E k1 ), the distance from the point E k2 to the vertex B k is L(B k E k2), L(B k E k1 )=0.1*L(A k B k ), L(B k E k2 )=0.5*L(A k B k ), the point H k1 Distance D from the vertex k The distance is L(D k H k1 ), the point H k2 Distance D from the vertex k The distance is L(D k H k2 ),L(D k H k1 )=0.1*L(A k D k ), L(D k H k2 )=0.5*L(A k D k ), the point E k1 and the point H k1 Forming boundary E k1 H k1 , the point E k2 and the point H k2 Forming boundary E k2 H k2 , the second sub-identification area b k2w Towards the boundary line A k B k and the boundary line A k D k The boundary line is located at the boundary line E k1 H k1 and the boundary E k2 H k2 between.

[0015] As a further improvement of an embodiment of the present invention, the second diagonal line B k D k Boundary line B on the other side k C k and boundary line D k C k Including point J k1 , click J k2 , click R k1 , click R k2 , the point J k1 and the point J k2 Located at the boundary line B k Ck above, the point R k1 and the point R k2 are located on the boundary line D k C k above, the point J k1 is at a distance of L(B k J k ) from the vertex B k1 , the point J k2 is at a distance of L(B k J k ) from the vertex B k2 , L(B k J k1 ) = 0.1 * L(B k C k ), L(B k J k2 ) = 0.5 * L(B k C k ), the point R k1 is at a distance of L(D k R k [[ID=4)) from the vertex D k1 , the point R k1 is at a distance of L(D k R k ) from the vertex D k2 , L(D k R k1 ) = 0.1 * L(D k C k ), L(D k R k2 ) = 0.5 * L(D k C k ), the point J k1 and the point R k1 form the boundary line J k1 R k1 , the point J k2 and the point R k2 form the boundary line J k2 R k2 , the second sub - identification area b k2w faces the boundary line B k C k and the boundary line D k C k and the boundary line between them is located between the boundary line J k1 R k1 and the boundary line J k2 R k2 .

[0016] As a further improvement of an embodiment of the present invention, wherein, n2 = n3, 3 ≤ n2 ≤ 5, 3 ≤ n3 ≤ 5, the first sub-identification area b k1u in the length direction includes boundary lines located on both sides of the first diagonal and parallel to the first diagonal, and the second sub-identification area b k2w in the length direction includes boundary lines located on both sides of the second diagonal and parallel to the second diagonal.

[0017] As a further improvement of an embodiment of the present invention, wherein, the first sub-identification area b k1u is symmetric with respect to the first diagonal, and the second sub-identification area b k2w is symmetric with respect to the second diagonal.

[0018] As a further improvement of an embodiment of the present invention, wherein, the endpoints of the boundary line of the first sub-identification area b k1u parallel to the first diagonal are located on the boundary line of the strip identification area b k , and for different first sub-identification areas b k1u located on the same boundary line of the strip identification area b k the interval between the endpoints is not less than one-tenth and not greater than one-fifth of the length of the boundary line where they are located;

[0019] The endpoints of the boundary line of the second sub-identification area b k2w parallel to the second diagonal are located on the boundary line of the strip identification area b k , and for different second sub-identification areas b k2w located on the same boundary line of the strip identification area b k the interval between the endpoints is not less than one-tenth and not greater than one-fifth of the length of the boundary line where they are located.

[0020] As a further improvement of an embodiment of the present invention, wherein, both the first sub-identification area b k1u and the second sub-identification area b k2w are rectangular regions, "obtaining the predicted variety type T k1u and confidence level C k1u of the first sub-identification area b k1u , obtaining the predicted variety type T k2w and confidence level C k2w of the second sub-identification area b k2w " includes:

[0021] respectively obtaining from the regional image formed by the strip identification area b k the first sub-identification area b k1uThe formed regional image and the second sub-identification area b k2w The formed regional image;

[0022] Respectively identify the first sub-identification area b k1u The formed regional image and the second sub-identification area b k2w The formed regional image, obtain the predicted variety type T k1u Of the first sub-identification area b k1u And confidence level C k1u And obtain the predicted variety type T k2w Of the second sub-identification area b k2w And confidence level C k2w .

[0023] As a further improvement of an embodiment of the present invention, wherein, "obtain the predicted variety type T k1u Of the first sub-identification area b k1u And confidence level C k1u , and obtain the predicted variety type T k2w Of the second sub-identification area b k2w And confidence level C k2w " includes:

[0024] Obtain the filled identification area image corresponding to each of the first sub-identification areas b k1u And the filled identification area image corresponding to each of the second sub-identification areas b k2w ,

[0025] The filled identification area image corresponding to the first sub-identification area b k1u Is: the image of the strip identification area b k After filling the area other than the first sub-identification area b k1u In the formed regional image with a unified pixel value; k Of the strip identification area b

[0026] The filled identification area image corresponding to the second sub-identification area b k2w Is: the image of the strip identification area b k After filling the area other than the second sub-identification area b k2w In the formed regional image with a unified pixel value; k Of the strip identification area b

[0027] Respectively identify the filled identification area image corresponding to the first sub-identification area b k1u And the filled identification area image corresponding to the second sub-identification area b k2w , and obtain the predicted variety type T k1u Of the first sub-identification area b k1u And confidence level Ck1u , obtain the second sub-identification area b k2w of the predicted item type T k2w and confidence level C k2w .

[0028] As a further improvement of an embodiment of the present invention, "input the item image into an image recognition model to obtain the strip recognition area b of the strip item k " includes:

[0029] Obtain the rectangular recognition area B generated by the image recognition model for each item in the item image P , the recognition area B P is a rectangular area generated by the image recognition model at the maximum contour line of each item, p = {1, 2,..., n4}, and n4 is the number of the recognition area B p ;

[0030] Identify the recognition area B p , and obtain the predicted item type T p and confidence level C p of each recognition area B p ;

[0031] Obtain the aspect ratio W p of the recognition frame B p where the confidence level C is between the first threshold and the second threshold, and the recognition area B p with the aspect ratio W p within the preset interval is the strip recognition area b p , and the first threshold is greater than the second threshold. k As a further improvement of an embodiment of the present invention, the first threshold is any value between 0.75 and 0.9, the second threshold is any value between 0.4 and 0.55, and the preset interval is [0.7, 1.4].

[0032] To achieve the above-mentioned invention purpose, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the image recognition method for strip items described in any of the above embodiments are implemented.

[0033] To achieve the above-mentioned invention purpose, an embodiment of the present invention provides an image recognition system, where the recognition 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, the steps in the image recognition method for strip items described in any of the above embodiments are implemented.

[0034] ​

[0035] To achieve the above-mentioned invention objective, an embodiment of the present invention provides a refrigeration device. The refrigeration device includes an image recognition 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, the steps in the image recognition method for strip-shaped items described in any of the above embodiments are implemented.

[0036] Compared with the prior art, by respectively obtaining a first sub-recognition area and a second sub-recognition area in the form of strips at two diagonal lines in the strip-shaped item recognition area, the beneficial effect of the present invention is that it can accurately identify strip-shaped items. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a schematic structural diagram of a refrigeration device according to an embodiment of the present invention;

[0038] Figure 2 is a schematic diagram of an item image and a recognition area according to an embodiment of the present invention;

[0039] Figure 3 is a schematic diagram of an image of a strip-shaped item recognition area, a first sub-recognition area, and a first sub-recognition area according to an embodiment of the present invention;

[0040] Figure 4 is a schematic diagram of an image of a strip-shaped item recognition area, a second sub-recognition area, and a corresponding filled recognition area of the second sub-recognition area according to an embodiment of the present invention;

[0041] Figure 5 is a flowchart of an image recognition method for strip-shaped items according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The present invention will be described in detail below with reference to the specific embodiments shown in the drawings. However, these embodiments do not limit the present invention, and any structural, method, or functional transformation made by those of ordinary skill in the art based on these embodiments is included in the protection scope of the present invention.

[0043] Refer to Figure 1 and Figure 2 , in an embodiment of the present invention, the refrigeration device 100 such as a refrigerator, a freezer, a commercial display cabinet, etc. may include a cabinet 1 and a door 2. A storage compartment for storing items 4 may be formed inside the cabinet 1, and the door 2 may be used to open and close the storage compartment. A bottle holder structure for storing items 4 may also be provided on the door 2, or a door compartment for storing items 4 may be formed inside the door 2, etc.

[0044] The refrigeration device 100 can be provided with an image recognition module 3. The image recognition module 3 can include an image recognition model and an image acquisition module. The image acquisition module can include a camera, and the camera can be used to acquire the item image 5 of the item 4 stored in the refrigeration device 100. The image recognition model can be used to perform object detection and category recognition on the item image 5. The refrigeration device 100 can learn information such as the category of the item stored in the refrigeration device 100 by obtaining the detection result output by the image recognition model, so as to manage the item 4 stored in the refrigeration device 100.

[0045] Referring to Figure 2 , in the process of processing the item image 5, the image recognition model can first perform object detection on the item image 5 and generate a rectangular recognition area covering the item 4 at the maximum contour of the target item in the image. The recognition area can be marked as B p . Where p = {1, 2,..., n4}, and n4 is the number of the recognition areas B p . A specific example is that if there are 3 recognition areas, the recognition areas can be B1, B2, and B3 respectively. The target item is the item 4 that the recognition area B p wants to recognize. The image recognition model can be an algorithm model constructed using a deep convolutional neural network, etc.

[0046] The recognition area B p is the area framed by the recognition box calculated and output by the image recognition model. The area position it contains can be identified by marking the upper left, lower right coordinates of the rectangular recognition area B p , that is, (x 左上 , y 左上 ) and (x 右下 , y 右下 ), or it can also be identified by marking the rectangular center coordinates and the width and height of the rectangle, that is, (x 中心 , y 中心 ), w 宽 and h 高 .

[0047] After generating the recognition area B p , the image recognition model can send each recognition area B p into the classification recognition network, or by extracting features from each recognition area B p and comparing the features with the pre-stored feature data, etc., to obtain the predicted item category and confidence corresponding to each recognition area B p .

[0048] The predicted item category can refer to the name information representing a specific item category predicted by the image recognition model, such as cucumber, carrot, etc.

[0049] The confidence level may be a value indicating the accuracy of the predicted item type. The confidence level may be a value between 0 and 1.

[0050] Reference Figures 2 - 5 Furthermore, in another embodiment of the present invention, an image recognition method for a strip-shaped article 41 is provided, which may include:

[0051] Acquire an item image 5 containing a strip-shaped item 41;

[0052] The object image 5 is input into the image recognition model to obtain the strip recognition area b of the strip object 41. k , k={1,2,…,n1}, n1 is the strip identification area b k The number of strips identifying the area b k The rectangular area generated by the image recognition model at the maximum contour line of the object 4, the rectangular strip recognition area b k may include intersecting first and second diagonals;

[0053] From the strip identification area b k Obtain the first sub-identification area b in the shape of a long strip k1u and the second sub-identification area b k2w ,u={1,2,…,n2}, n2 is the first sub-identification area b k1u The number of w={1,2,…,n3}, n3 is the second sub-identification area b k2w The number of the first sub-identification area b k1u The length direction of extends along the first diagonal line, and the first diagonal line is at least partially located in the first sub-identification area b k1u In the second sub-identification area b k2w The length direction of the second diagonal line extends along the second diagonal line, and the second diagonal line is at least partially located in the second sub-identification area b k2w Inside;

[0054] Get the first sub-identification area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w ;

[0055] From the confidence level C k1u and confidence C k2w Get the maximum confidence ZC k , if the maximum confidence ZC k is greater than the first threshold, the strip identification area bk The type of the output item is the maximum confidence level ZC k The corresponding predicted item type.

[0056] In this embodiment, the strip-shaped item 41 may refer to an item 4 whose length is twice or more than twice its width, such as a cucumber, a strip-shaped beverage, etc. The strip-shaped item recognition area may refer to the recognition area B corresponding to the strip-shaped item 41 p Marked as b k , k = {1, 2,..., n1}, and n1 may be the number of strip-shaped item recognition areas b p in the recognition area B k Specifically, for example, if there are 3 strip-shaped item recognition areas, they can be b1, b2, and b3 respectively. There may be multiple strip-shaped items 41 in the item image 5, so the strip-shaped item recognition area b k can also be multiple. Each rectangular strip-shaped item recognition area b k may include two intersecting diagonals, and the diagonals may include a first diagonal and a second diagonal.

[0057] The first threshold may be a certain value preset as a confidence level comparison benchmark. If the confidence level is greater than the first threshold, it means that the confidence level is relatively high, and the predicted item type corresponding to this confidence level is very likely to be the item type of the target item corresponding to the recognition area B p The predicted item type corresponding to this confidence level can be used as the item type of the output for the recognition area B p If the confidence level is less than the first threshold, it means that the confidence level is relatively low, and the predicted item type corresponding to this confidence level may be inaccurate. Therefore, the recognition area B p may not output the predicted item type information.

[0058] In this embodiment, the sub-recognition area 8 may refer to an area within the strip-shaped item recognition area b k for individual recognition. The area of the sub-recognition area 8 is smaller than the corresponding strip-shaped item recognition area b k , and several sub-recognition areas 8 may be included within the same strip-shaped item recognition area b k . The areas covered by the sub-recognition areas 8 within the same strip-shaped item recognition area b k are not the same, but may partially overlap.

[0059] The sub-recognition area 8 may include a first sub-recognition area b k1u and a second sub-recognition area b k2w . The first sub-recognition area b k1u ,

[0060] The first sub-recognition area b k1u may refer to the strip-shaped item recognition area b kA bar-shaped area for individual identification extending along the first diagonal in the length direction. u = {1, 2, …, n2}, where n2 is the number of the first sub-identification areas b k1u . For a specific example, if there are 3 first sub-identification areas, they can be b k11 , b k12 , b k13 respectively. The predicted variety type of the first sub-identification area b k1u can be marked as T k1u , and the confidence level can be marked as C k1u .

[0061] The second sub-identification area b k2w can refer to a sub-identification area 8 extending along the second diagonal in the length direction within the strip identification area b k . w = {1, 2, …, n3}, where n3 is the number of the second sub-identification areas b k2w . For a specific example, if there are 3 strip identification areas, they can be b k21 , b k22 , b k23 respectively. The predicted variety type of the second sub-identification area b k2w can be marked as T k2w , and the confidence level can be marked as C k2w ;

[0062] In the actual identification process, since the length and width of the strip-shaped item 41 differ greatly, within the rectangular identification area B p generated along its maximum contour line, there may be other items 4 included. If the strip identification area b k is directly identified as a whole, it often leads to errors in the predicted variety type of the strip identification area b k , a relatively low confidence level of the predicted variety type, etc.

[0063] By selecting a number of first sub-identification areas b k and second sub-identification areas b k1u at the diagonal in the strip identification area b k2w , it is ensured that the image information of the strip-shaped item 41 can be effectively selected. The first sub-identification area b k1u and the second sub-identification area b k2w are bar-shaped and extend along the diagonal direction in the length direction, which can ensure that more and more comprehensive item image 5 information of the selected strip-shaped item 41 is obtained. At the same time, the interference of the image information of other items 4 is reduced.

[0064] Since the item image 5 information covered in the sub-identification area 8 is relative to the strip identification area b kTherefore, when the sub-recognition areas 8 are identified separately, the confidence level of the predicted item type is relatively high, and the predicted item type is relatively accurate. Since different sub-recognition areas 8 cover different areas, the corresponding predicted item types may be the same or different. Therefore, the predicted item type corresponding to the maximum confidence level greater than the first threshold can be selected as the bar identification area b. k The type of item being output.

[0065] Adopt the design scheme of this embodiment, by k The first sub-identification area b of the bar is obtained at the two diagonal lines in k1u and the second sub-identification area b k2w , can effectively obtain more and more comprehensive image information of the strip-shaped article 41, and at the same time reduce the interference of the image information of other articles 4, by identifying the first sub-identification area b k1u and the second sub-identification area b k2w Able to realize strip identification area b k The corresponding strip-shaped article 41 is accurately identified, ensuring the accuracy of the identification of the strip-shaped article 41 and preventing the incorrect identification and missed identification of the strip-shaped article 41.

[0066] Reference Figure 3 Furthermore, in another embodiment of the present invention, the image recognition method for the strip-shaped article 41, wherein the rectangular strip-shaped article recognition area b k The four vertices are A k 、B k 、C k 、D k , the first diagonal is A k C k , the second diagonal is B k D k .

[0067] Located on the first diagonal A k C k Boundary line A on one side k B k and boundary line B k C k Can include point M k1 、Point M k2 、Point N k1 、Point N k2 , the point M k1 and the point M k2 Located on the boundary line A k B k On the point N k1 and the point N k2 Located at the boundary line Bk C k superior.

[0068] The point M k1 From the vertex A k The distance is L(A k M k1 ), the point M k2 From the vertex A k2 The distance is L(A k M k2 ), the boundary line A k B k The length is L(A k B k ).

[0069] L(A k M k1 )=0.1*L(A k B k ), L(A k M k2 )=0.5*L(A k B k ).

[0070] The point N k1 From the vertex C k The distance is L(C k N k1 ), the point N k2 From the vertex C k The distance is L(C k N k2 ), the boundary line B k C k The length is L(B k C k ).

[0071] L(C k N k1 )=0.1*L(B k C k ), L(C k N k2 )=0.5*L(B k C k ).

[0072] The point M k1 and the point N k1 Forming boundary M k1 N k1 , the point M k2 and the point N k2 Forming boundary M k2 N k2 , the first sub-identification area bk1u Towards the boundary line A k B k and the boundary line B k C k The boundary line of and is located at the boundary line M k1 N k1 and the boundary line M k2 N k2 Between.

[0073] Since the area covered by the sub-identification area 8 determines the accuracy of the final identification result of the sub-identification area 8. If the selected area of the sub-identification area 8 is too small, it may lead to too little characteristic information of the item 4 being obtained, resulting in inaccurate identification. If the selected area of the sub-identification area 8 is too large, it may lead to too much image information of other items 4 outside the strip-shaped item 41 being covered by the sub-identification area 8, resulting in inaccurate identification results.

[0074] Therefore, adopting the design scheme of this embodiment can ensure that the sub-identification area 8 can cover an appropriate area range, prevent situations such as too little image information of the item included in the sub-identification frame and incomplete image information covered due to the area range covered by the sub-identification area 8 being too small, and at the same time avoid too much image information of other items 4 being included in the sub-identification area 8 due to the selected area range of the sub-identification area 8 being too large, resulting in a decrease in the accuracy of the identification result of the sub-identification area 8.

[0075] Refer to Figure 3 , further, in another embodiment of the present invention, the image recognition method for the strip-shaped item 41, wherein, the boundary line A on the other side of the first diagonal A k C k The boundary line A k D k And the boundary line D k C k May include point P k1 , point P k2 , point Q k1 , point Q k2 , the point P k1 And the point P k2 Are located on the boundary line A k D k The point Q k1 And the point Q k2 Are located on the boundary line D k C k .

[0076] The point P k1 The distance from the vertex A k Is L(A k P k1 ), the point P k2The distance from the vertex A k is L(A k P k2 ), and the length of the boundary line A k D k is L(A k D k ).

[0077] L(A k P k1 ) = 0.1 * L(A k D k ), L(A k P k2 ) = 0.5 * L(A k D k ).

[0078] The point Q k1 is at a distance of L(C k from the vertex C k Q k1 ), and the point Q k2 is at a distance of L(C k from the vertex C k Q k2 ), and the length of the boundary line D k C k is L(D k C k ).

[0079] L(C k Q k1 ) = 0.1 * L(D k C k ), L(C k Q k2 ) = 0.5 * L(D k C k ).

[0080] The point P k1 and the point Q k1 form a boundary line P k1 Q k1 ; the point P k2 and the point Q k2 form a boundary line P k2 Q k2 ; the boundary line of the first sub - recognition area b k1u faces the boundary line A k D k and the boundary line D k C k and is located between the boundary lines P k1 Q k1 and the boundary lines P k2 Qk2 between

[0081] Since the area covered by the sub-identification area 8 determines the accuracy of the final identification result of the sub-identification area 8. If the selected area of the sub-identification area 8 is relatively off, for example, only on one side of the first diagonal, it may also lead to too little characteristic information of the item 4 being obtained and unable to be accurately identified.

[0082] With such a setting, it can be ensured that the sub-identification area 8 can cover an appropriate area range, and at the same time, the area range covered by the selected sub-identification area 8 can be made more uniform and reasonable, preventing the situation where the sub-identification area 8 only selects the area on one side of the diagonal, ensuring that the image information covered by the sub-identification area 8 is more comprehensive, and thus ensuring the accuracy of the identification result of the sub-identification area 8.

[0083] Refer to Figure 4 , further, in another embodiment of the present invention, the image recognition method for the strip-shaped item 41, wherein, the boundary line A k D k on one side of the second diagonal B k B k and the boundary line A k D k may include the point E k1 , the point E k2 , the point H k1 , the point H k2 , the point E k1 and the point E k2 are located on the boundary line A k B k )], the point H k1 and the point H k2 are located on the boundary line A k D k .

[0084] The distance from the point E k1 to the vertex B k is L(B k E k1 ), the distance from the point E k2 to the vertex B k is L(B k E k2 ), L(B k E k1 ) = 0.1 * L(A k B k ), L(B k E k2 ) = 0.5 * L(A k B k ).

[0085] The point Hk1 The distance from the vertex D k is L(D k H k1 ), and the point H k2 is at a distance of L(D k from the vertex D k H k2 ), L(D k H k1 ) = 0.1 * L(A k D k ), L(D k H k2 ) = 0.5 * L(A k D k ).

[0086] The point E k1 and the point H k1 form the boundary line E k1 H k1 ; the point E k2 and the point H k2 form the boundary line E k2 H k2 ; the second sub - recognition area b k2w faces the boundary line A k B k and the boundary line A k D k 's boundary line is located between the boundary line E k1 H k1 and the boundary line E k2 H k2 .

[0087] With such a setting, it can be ensured that the sub - recognition area 8 can cover an appropriate area range, preventing situations such as too little information of the item image 5 contained in the sub - recognition frame and incomplete image information covered due to the sub - recognition area 8 covering too small an area range. At the same time, it can avoid the situation that the sub - recognition area 8 contains too much image information of other items 4 due to the selected area range of the sub - recognition area 8 being too large, resulting in a reduction in the accuracy of the recognition result of the sub - recognition area 8.

[0088] Referring to Figure 4 , further, in another embodiment of the present invention, for the image recognition method of the strip - shaped item 41, where

[0089] The boundary line B k D k on the other side of the second diagonal B k C k and the boundary line D k C k may include the point J k1 and the point Jk2 , point R k1 , point R k2 , the point J k1 and the point J k2 are located on the boundary line B k C k ), the point R k1 and the point R k2 are located on the boundary line D k C k ),

[0090] The distance from the point J k1 to the vertex B k is L(B k J k1 ), the distance from the point J k2 to the vertex B k is L(B k J k2 ), L(B k J k1 ) = 0.1 * L(B k C k ), L(B k J k2 ) = 0.5 * L(B k C k ),

[0091] The distance from the point R k1 to the vertex D k is L(D k R k1 ), the distance from the point R k1 to the vertex D k is L(D k R k2 ), L(D k R k1 ) = 0.1 * L(D k C k ), L(D k R k2 ) = 0.5 * L(D k C k ),

[0092] The point J k1 and the point R k1 form the boundary line J k1 R k1 ), the point J k2 and the point R k2 form the boundary line J k2 R k2 ), the second sub - identification area b k2w faces the boundary line Bk C k and the boundary line D k C k The boundary line of is located at the boundary line J k1 R k1 and the boundary line J k2 R k2 between them.

[0093] With such a setting, it can be ensured that the sub-identification area 8 can cover an appropriate area range, and the area range covered by the selected sub-identification area 8 can be made more uniform and reasonable, preventing the situation where the sub-identification area 8 only selects the area on one side of the diagonal, ensuring that the image information covered by the sub-identification area 8 is more comprehensive, and thus ensuring the accuracy of the recognition result of the sub-identification area 8.

[0094] Furthermore, in another embodiment of the present invention, in the image recognition method for the strip-shaped article 41, n2 = n3, 3 ≤ n2 ≤ 5, 3 ≤ n3 ≤ 5.

[0095] With such a setting, it can be ensured that an equal number of sub-identification areas 8 are selected at the two diagonals of the strip-shaped object recognition area b k , and at the same time, it is ensured that the number of the first sub-identification area b k1u and the second sub-identification area b k2w is appropriate, avoiding the situation that the covered image information is not comprehensive due to too few selected sub-identification areas 8, and further resulting in inaccurate recognition results of the strip-shaped object recognition area b k , and avoiding excessive calculation amount and long calculation time due to too many selected sub-identification areas 8, resulting in low recognition efficiency of the image recognition model.

[0096] Referring to Figure 3 and Figure 4 , furthermore, in another embodiment of the present invention, the length direction of the first sub-identification area b k1u may include boundary lines located on both sides of the first diagonal and parallel to the first diagonal. The length direction of the second sub-identification area b k2w may include boundary lines located on both sides of the second diagonal and parallel to the second diagonal.

[0097] With such a setting, it can be ensured that the length direction of the sub-identification area 8 extends along the diagonal direction, and it is also ensured that the areas at different positions on the same side of the diagonal selected by the sub-identification area 8 are more uniform, and the area range covered by the sub-identification area 8 can be made more uniform and reasonable, thereby ensuring the accuracy of the recognition result of the sub-identification area 8.

[0098] Referring to Figure 3 and Figure 4, Further, in another embodiment of the present invention, the image recognition method for the strip-shaped article 41, wherein the first sub-recognition area b k1u is symmetric with respect to the first diagonal. The second sub-recognition area b k2w is symmetric with respect to the second diagonal.

[0099] With such a setting, it can be ensured that the length direction of the sub-recognition area 8 extends along the diagonal direction, and it is also ensured that the areas at the same positions on both sides of the diagonal selected by the sub-recognition area 8 are more uniform, enabling the area range covered by the sub-recognition area 8 to be more uniform and reasonable, thereby ensuring the accuracy of the recognition result of the sub-recognition area 8.

[0100] Refer to Figure 3 and Figure 4 , Further, in another embodiment of the present invention, the image recognition method for the strip-shaped article 41, wherein the first sub-recognition area b k1u The endpoints of the boundary line parallel to the first diagonal are located on the boundary line of the strip-shaped article recognition area b k , and for different first sub-recognition areas b k1u The intervals between the endpoints located on the same boundary line of the strip-shaped article recognition area b k are not less than one-tenth of the length of the boundary line where they are located and not greater than one-fifth of the length of the boundary line where they are located.

[0101] For the second sub-recognition area b k2w The endpoints of the boundary line parallel to the second diagonal are located on the boundary line of the strip-shaped article recognition area b k , and for different second sub-recognition areas b k2w The intervals between the endpoints located on the same boundary line of the strip-shaped article recognition area b k are not less than one-tenth of the length of the boundary line where they are located and not greater than one-fifth of the length of the boundary line where they are located.

[0102] In this embodiment, the sub-recognition area 8 can extend as far as possible along the diagonal towards both ends of the diagonal on the premise of not exceeding the boundary line of the strip-shaped article recognition area b k . The endpoints of the boundary line of the sub-recognition area 8 parallel to the diagonal can be located on the boundary line of the strip-shaped article recognition area b k . One of the endpoints of the boundary line of the sub-recognition area 8 parallel to the diagonal can be located on the boundary line of the strip-shaped article recognition area b k , or both endpoints can be located on the boundary line of the strip-shaped article recognition area b k .

[0103] With such a setting, the area covered by the selected sub-identification area 8 can be made more uniform and reasonable, while ensuring that the image information covered by the sub-identification area 8 is more comprehensive, thereby ensuring the accuracy of the recognition result of the sub-identification area 8.

[0104] Different ones of the first sub-identification areas b k1u Located in the strip identification area b k The endpoints on the same boundary line can be spaced a certain distance apart. Different ones of the first sub-identification areas b k1u The boundary lines can be spaced a certain distance apart. Different ones of the second sub-identification areas b k2w Located in the strip identification area b k The endpoints on the same boundary line can be spaced a certain distance apart. Different ones of the second sub-identification areas b k2w The boundary lines can be spaced a certain distance apart.

[0105] With such a setting, it can ensure the diversity of the image information of the item 4 covered by different sub-identification areas 8, prevent the situation where multiple sub-identification areas 8 repeatedly select a certain specific area, make the image information covered by the set of sub-identification areas 8 more comprehensive, and the position distribution of different sub-identification areas 8 more reasonable, thereby ensuring the accuracy of the recognition result of the strip identification area b k of.

[0106] Refer to Figure 3 , further, in another embodiment of the present invention, the image recognition method for the strip item 41, wherein, the first sub-identification area b k1u and the second sub-identification area b k2w can both be rectangular areas.

[0107] "Obtain the predicted item type T k1u and confidence level C k1u of the first sub-identification area b k1u , obtain the predicted item type T k2w and confidence level C k2w of the second sub-identification area b k2w " can include:

[0108] Respectively obtain the area image formed by the first sub-identification area b k and the area image formed by the second sub-identification area b k1u from the area image formed by the strip identification area b k2w ;

[0109] Respectively identify the area image formed by the first sub-identification area b k1u and the area image formed by the second sub-identification area b k2w , obtain the first sub-identification area bk1u The predicted item type T k1u and the confidence level C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and the confidence level C k2w .

[0110] In this embodiment, the image of the strip-shaped object recognition area b k The regional image formed by can be cropped to obtain the images separately formed by the sub-identification area 8, and then the images formed by the sub-identification area 8 can be separately recognized to obtain the item type information and confidence level corresponding to each sub-identification area 8.

[0111] The sub-identification areas 8 are all rectangular areas, which is convenient for cropping operations, ensuring that the images formed by the individual sub-identification areas 8 can be directly and quickly obtained, reducing the calculation amount and improving the calculation efficiency.

[0112] With such a setting, it is possible to separately obtain the images separately formed by each sub-identification area 8 from the regional image formed by the strip-shaped object recognition area b k The images formed by each sub-identification area 8 can be separately recognized to obtain the item type information and confidence level corresponding to each sub-identification area 8, with accurate recognition, convenience, speed, and high efficiency.

[0113] Refer to Figure 4 , further, in another embodiment of the present invention, the image recognition method for the strip-shaped item 41, wherein, "obtain the predicted item type T k1u of the first sub-identification area b k1u and the confidence level C k1u , obtain the predicted item type T k2w of the second sub-identification area b k2w and the confidence level C k2w " may include:

[0114] Obtain the filled recognition area image 10 corresponding to each of the first sub-identification areas b k1u and the filled recognition area image 10 corresponding to each of the second sub-identification areas b k2w ,

[0115] The filled recognition area image 10 corresponding to the first sub-identification area b k1u is: the image of the strip-shaped object recognition area b k formed by filling the area other than the first sub-identification area b k1u with a unified pixel value in the formed regional image; k The image of the strip-shaped object recognition area b

[0116] The filled recognition area image 10 corresponding to the second sub-identification area b k2wThe corresponding filled identification area image 10 is: the strip identification area b k The formed area image is excluding the second sub-identification area b k2w The strip identification area b is obtained by filling the area outside with a uniform pixel value k images;

[0117] Respectively identify the first sub-identification area b k1u The corresponding filled recognition area image 10 and the second sub-recognition area b k2w The corresponding filled recognition area image 10 is obtained by k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w .

[0118] In this embodiment, by k The formed area image is excluding the second sub-identification area b k2w The areas outside the sub-recognition area 8 are filled with a uniform pixel value, for example, a uniform pixel value of (127, 127, 127). This can erase image information other than that covered by the sub-recognition area 8, preventing the areas outside the sub-recognition area 8 from interfering with the individual recognition of the sub-recognition area 8. The graphical information of the object 4 covered by each sub-recognition area 8 can then be recognized by identifying the filled-in area image 10 corresponding to each sub-recognition area 8, thereby obtaining the object type information and confidence level corresponding to each sub-recognition area 8.

[0119] Furthermore, since the recognition of the image information covered by the sub-recognition area 8 is achieved by recognizing the filled recognition area image corresponding to the sub-recognition area 8, the shape of the sub-recognition area 8 may be irregular.

[0120] With such a setting, it is possible to obtain the filled recognition area image 10 corresponding to each sub-recognition area 8, to recognize irregularly shaped sub-recognition areas 8, to prevent areas outside the sub-recognition area 8 from interfering with the individual recognition of the sub-recognition area 8, and to recognize the object image 5 information covered by each sub-recognition area 8 to obtain the object type information and confidence level corresponding to each sub-recognition area 8, with accurate, convenient, fast and efficient recognition.

[0121] Reference Figure 2 Furthermore, in another embodiment of the present invention, the image recognition method for the strip-shaped article 41, wherein, "the article image 5 is input into the image recognition model to obtain the strip recognition area b of the strip-shaped article 41 k ” can include:

[0122] Obtain the rectangular recognition area B generated by the image recognition model for each object 4 in the object image 5 p , the identification area B p Generate a rectangular area at the maximum contour line of each object 4 for the image recognition model, p = {1, 2, ..., n4}, n4 is the recognition area B p The number of

[0123] Identify the identification area B p , and obtain each of the identification areas B p The predicted item type T p and confidence C p ;

[0124] Get the confidence C p The identification box B between the first threshold and the second threshold p The width-to-length ratio W p , the aspect ratio W p The identification area B in the preset interval p The strip identification area b k , the first threshold is greater than the second threshold.

[0125] In this embodiment, the aspect ratio refers to the rectangular identification area B p The ratio of two adjacent boundary lines.

[0126] In this embodiment, each recognition area B generated by the image recognition model can be p Perform overall recognition and obtain each identification area B p The predicted item types and confidence levels are calculated, and some identification areas B are preliminarily screened out based on the confidence levels, where the confidence levels do not meet the requirements and the corresponding predicted item types are uncertain whether they are correct. p Perform secondary identification.

[0127] If the confidence level is greater than the first threshold, it means that the recognition area B corresponding to the confidence level p The predicted item type is likely to be correct and can be used as the identification area B p If the confidence level is less than the second threshold, it means that the confidence level is too low, and the recognition area B corresponding to the confidence level is p It is likely to be wrong. The recognition frame may not correctly cover the target object, such as recognition area B p It may include containers containing articles 4, so for this identification area B p The item type does not need to be output and secondary identification is not required.

[0128] And because only when the strip-shaped article 41 is in the identification area B pWhen the inner inclination angle is relatively large, that is, recognition area B p When it is close to a square, the recognition area B where it is located p covers more information of other item 4, which may cause misrecognition of the strip-shaped item 41. If the recognition area B p itself is long and strip-shaped, that is, the aspect ratio value is low or the aspect ratio value is high, it can be considered that the recognition area B p fits well with the shape of the strip-shaped item 41. Even if the recognition area B p contains information of other item 4, it is not enough to have a significant impact on the final recognition result. Therefore, the operation of obtaining the secondary recognition of the above-mentioned sub-recognition area 8 can be not performed.

[0129] Set like this, it is possible to screen out the strip-shaped item recognition area b that needs to perform the operation of obtaining the sub-recognition area 8 for secondary recognition through the confidence level and aspect ratio k , avoiding performing the secondary recognition operation of obtaining the sub-recognition area 8 for each recognition area B p can reduce the calculation amount, reduce the calculation time, improve the recognition efficiency of the image recognition model, and at the same time ensure the accuracy and reliability of the recognition result, avoiding misrecognition and missed recognition.

[0130] Furthermore, in another embodiment of the present invention, in the image recognition method for the strip-shaped item 41, wherein, the first threshold is any value between 0.75 and 0.9, and the second threshold is any value between 0.4 and 0.55.

[0131] The preset interval is [0.7, 1.4].

[0132] Set like this, it is possible to screen out the strip-shaped item recognition area b k , avoiding performing the secondary recognition operation of obtaining the sub-recognition area 8 for each recognition area B p can reduce the calculation amount, reduce the calculation time, improve the recognition efficiency of the image recognition model, and at the same time ensure the accuracy and reliability of the recognition result, avoiding misrecognition and missed recognition.

[0133] In one embodiment of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored. Among them, when the computer program is executed by a processor, the steps in the image recognition method for the strip-shaped item 41 described in any of the above embodiments are implemented.

[0134] In one embodiment of the present invention, there is provided an image recognition system. Among them, the recognition system may include a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, the steps in the image recognition method for the strip-shaped item 41 described in any of the above embodiments are implemented.

[0135] In an embodiment of the present invention, a refrigeration device 100 is provided. The refrigeration device 100 may include an image recognition system, and the image recognition system may include a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, the steps in the image recognition method for the strip-shaped article 41 described in any of the above embodiments are implemented.

[0136] In summary, an image recognition method, a storage medium, a system and a device for a strip-shaped article 41 in the present invention can respectively obtain a first sub-recognition area b k and a second sub-recognition area b k1u which are strip-shaped at two diagonal lines in the strip-shaped article recognition area b k2w , so as to solve the problem of incorrect recognition of the strip-shaped article 41 caused by the direct recognition method for the strip-shaped article 41 in the prior art.

[0137] By adopting the technical solution in the present application, accurate recognition of the corresponding strip-shaped article 41 in the strip-shaped article recognition area b k can be achieved, ensuring the accuracy of the recognition of the strip-shaped article 41, preventing incorrect recognition and missed recognition of the strip-shaped article 41. At the same time, the area range covered by the selected sub-recognition area 8 can be made more uniform and reasonable, ensuring that the image information covered by the sub-recognition area 8 is more comprehensive, ensuring the accuracy of the recognition result of the sub-recognition area 8. In addition, the calculation amount can be reduced, the calculation time can be shortened, and the recognition efficiency can be improved.

[0138] It should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and 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.

[0139] The series of detailed descriptions listed above are only specific descriptions of the feasible embodiments of the present invention, and they are not used to limit the protection scope of the present invention. Any equivalent embodiments or modifications made without departing from the technical spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for image recognition of strip-shaped articles, characterized in that: include: Get the item image containing the strip item; Input the object image into the image recognition model to obtain the strip object recognition area b of the strip object k , k={1,2,…,n1}, n1 is the number of the strip identification area, the strip identification area b k It is a rectangular area generated by the image recognition model at the maximum outline of the object, and the rectangular strip recognition area includes a first diagonal line and a second diagonal line intersecting each other; From the strip identification area b k Obtain the first sub-identification area b in the shape of a long strip k1u and the second sub-identification area b k2w ,u={1,2,…,n2}, n2 is the first sub-identification area b k1u The number of w={1,2,…,n3}, n3 is the second sub-identification area b k2w The number of The first sub-identification area b k1u The length direction of extends along the first diagonal line, and the first diagonal line is at least partially located in the first sub-identification area b k1u Inside, The second sub-identification area b k2w The length direction of the second diagonal line extends along the second diagonal line, and the second diagonal line is at least partially located in the second sub-identification area b k2w Inside; Get the first sub-identification area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w ; From the confidence level C k1u and confidence C k2w Get the maximum confidence ZC k , if the maximum confidence ZC k is greater than the first threshold, the strip identification area b k The output item type is the maximum confidence ZC k The corresponding predicted item type.

2. The image recognition method for strip-shaped articles according to claim 1, characterized in that: The rectangular strip identification area b k The four vertices are A k 、B k 、C k 、D k , the first diagonal is A k C k , the second diagonal is B k D k , Located on the first diagonal A k C k Boundary line A on one side k B k and boundary line B k C k Including point M k1 、Point M k2 、Point N k1 、Point N k2 , the point M k1 and the point M k2 Located on the boundary line A k B k On the point N k1 and the point N k2 Located at the boundary line B k C k superior, The point M k1 From the vertex A k The distance is L(A k M k ), the point M k2 From the vertex A k2 The distance is L(A k M k2 ), the boundary line A k B k The length is L(A k B k ), L(A k M k1 )=0.1*L(A k B k ), L(A k M k2 )=0.5*L(A k B k ), The point N k1 From the vertex C k The distance is L(C k N k1 ), the point N k2 From the vertex C k The distance is L(C k N k2 ), the boundary line B k C k The length is L(B k C k ), L(C k N k1 )=0.1*L(B k C k ), L(C k N k2 )=0.5*L(B k C k ), The point M k1 and the point N k1 Forming boundary M k1 N k1 , the point M k2 and the point N k2 Forming boundary M k2 N k2 , the first sub-identification area b k1u Towards the boundary line A k B k and the boundary line B k C k The boundary line is located at the boundary line M k1 N k1 and the boundary M k2 N k2 between.

3. The image recognition method for strip-shaped articles according to claim 2, characterized in that: Located on the first diagonal A k C k Boundary line A on the other side k D k and boundary line D k C k Including point P k1 , click P k2 , click Q k1 , click Q k2 , the point P k1 and the point P k2 Located on the boundary line A k D k On the point Q k1 and the point Q k2 Located on the boundary line D k C k superior, The point P k1 From the vertex A k The distance is L(A k P k1 ), the point P k2 From the vertex A k The distance is L(A k P k2 ), the boundary line A k D k The length is L(A k D k ), L(A k P k1 )=0.1*L(A k D k ), L(A k P k2 )=0.5*L(A k D k ), The point Q k1 From the vertex C k The distance is L(C k Q k1 ), the point Q k2 From the vertex C k The distance is L(C k Q k2 ), the boundary line D k C k The length is L(D k C k ), L(C k Q k1 )=0.1*L(D k C k ), L(C k Q k2 )=0.5*L(D k C k ), The point P k1 and the point Q k1 Forming boundary P k1 Q k1 , the point P k2 and the point Q k2 Forming boundary P k2 Q k2 , the first sub-identification area b k1u Towards the boundary line A k D k and the boundary line D k C k The boundary line is located at the boundary line P k1 Q k1 and the boundary line P k2 Q k2 between.

4. The image recognition method for strip-shaped articles according to claim 3, characterized in that: Located on the second diagonal B k D k Boundary line A on one side k B k and boundary line A k D k Including point E k1 , point E k2 、Point H k1 、Point H k2 , the point E k1 and the point E k2 Located on the boundary line A k B k On the point H k1 and the point H k2 Located on the boundary line A k D k superior, The point E k1 From the vertex B k The distance is L(B k E k1 ), the point E k2 From the vertex B k The distance is L(B k E k2 ), L(B k E k1 )=0.1*L(A k B k ), L(B k E k2 )=0.5*L(A k B k ), The point H k1 Distance D from the vertex k The distance is L(D k H k1 ), the point H k2 Distance D from the vertex k The distance is L(D k H k2 ),L(D k H k1 )=0.1*L(A k D k ), L(D k H k2 )=0.5*L(A k D k ), The point E k1 and the point H k1 Forming boundary E k1 H k1 , the point E k2 and the point H k2 Forming boundary E k2 H k2 , the second sub-identification area b k2w Towards the boundary line A k B k and the boundary line A k D k The boundary line is located at the boundary line E k1 H k1 and the boundary E k2 H k2 between.

5. The image recognition method for strip-shaped articles according to claim 4, characterized in that: Located on the second diagonal B k D k Boundary line B on the other side k C k and boundary line D k C k Including point J k1 , click J k2 , click R k1 , click R k2 , the point J k1 and the point J k2 Located at the boundary line B k C k On the point R k1 and the point R k2 Located on the boundary line D k C k superior, Point J k1 From the vertex B k The distance is L(B k J k1 ), the point J k2 From the vertex B k The distance is L(B k J k2 ),L(B k J k1 )=0.1*L(B k C k ), L(B k J k2 )=0.5*L(B k C k ), The point R k1 Distance D from the vertex k The distance is L(D k R k1 ), the point R k1 Distance D from the vertex k The distance is L(D k R k2 ), L(D k R k1 )=0.1*L(D k C k ), L(D k R k2 )=0.5*L(D k C k ), Point J k1 and the point R k1 Forming Boundary J k1 R k1 , the point J k2 and the point R k2 Forming Boundary J k2 R k2 , the second sub-identification area b k2w Towards the boundary line B k C k and the boundary line D k C k The boundary line is located at the boundary line J k1 R k1 and the boundary J k2 R k2 between.

6. The image recognition method for strip-shaped articles according to claim 5, characterized in that: n2=n3,3≤n2≤5,3≤n3≤5, The first sub-identification area b k1u The length direction includes boundary lines located on both sides of the first diagonal line and parallel to the first diagonal line, and the second sub-identification area b k2w The length direction includes boundary lines respectively located on both sides of the second diagonal line and parallel to the second diagonal line.

7. The image recognition method for strip-shaped articles according to claim 6, characterized in that: The first sub-identification area b k1u Symmetrically with respect to the first diagonal line, the second sub-identification area b k2w Symmetrical with respect to the second diagonal line.

8. The image recognition method for strip-shaped articles according to claim 6, characterized in that: The first sub-identification area b k1u The endpoint of the boundary line parallel to the first diagonal line is located in the strip identification area b k On the boundary line of the first sub-identification area b, and different k1u Located in the strip identification area b k The distance between the endpoints on the same boundary line is not less than one tenth and not more than one fifth of the length of the boundary line; The second sub-identification area b k2w The endpoint of the boundary line parallel to the second diagonal line is located in the strip identification area b k On the boundary line of the second sub-identification area b, and different k2w Located in the strip identification area b k The interval between each of the endpoints on the same boundary line is not less than one tenth of the length of the boundary line and not more than one fifth of the length of the boundary line.

9. The image recognition method for strip-shaped articles according to claim 1, characterized in that: The first sub-identification area b k1u and the second sub-identification area b k2w are all rectangular areas, "Get the first sub-identification area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w "include: From the strip identification area b k The first sub-identification area b is obtained from the formed regional image. k1u The formed regional image and the second sub-identification area b k2w The formed regional image; Respectively identify the first sub-identification area b k1u The formed regional image and the second sub-identification area b k2w The formed area image is obtained to obtain the first sub-identification area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w .

10. The image recognition method for strip-shaped articles according to claim 1, characterized in that: "Get the first sub-identification area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w "include: Get each of the first sub-identification areas b k1u The corresponding filled identification area image and each of the second sub-identification areas b k2w The corresponding filled recognition area image, The first sub-identification area b k1u The corresponding filling identification area image is: the strip identification area b k The formed area image is excluding the first sub-identification area b k1u The strip identification area b is obtained by filling the area outside with a uniform pixel value k images; The second sub-identification area b k2w The corresponding filling identification area image is: the strip identification area b k The formed area image is excluding the second sub-identification area b k2w The strip identification area b is obtained by filling the area outside with a uniform pixel value. k images; Respectively identify the first sub-identification area b k1u The corresponding filled recognition area image and the second sub-recognition area b k2w The corresponding filled recognition area image is obtained to obtain the first sub-recognition area b k1u The predicted item type T k1u and confidence C k1u , obtain the second sub-identification area b k2w The predicted item type T k2w and confidence C k2w .

11. The image recognition method for strip-shaped articles according to claim 1, characterized in that: Input the item image into the image recognition model to obtain the strip identification area b of the strip item k "include: Obtain the rectangular recognition area B generated by the image recognition model for each object in the object image P , the identification area B P Generate a rectangular area at the maximum contour line of each object for the image recognition model, p = {1, 2, ..., n4}, n4 is the recognition area B p The number of Identify the identification area B p , and obtain each of the identification areas B p The predicted item type T p and confidence C p ; Get the confidence C p The identification area B between the first threshold and the second threshold p The width-to-length ratio W p , the aspect ratio W p The identification area B in the preset interval p The strip identification area b k , the first threshold is greater than the second threshold.

12. The image recognition method for strip-shaped articles according to claim 11, characterized in that: The first threshold is any value between 0.75 and 0.9, the second threshold is any value between 0.4 and 0.55, and the preset interval is [0.7, 1.4].

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the image recognition method for strip-shaped articles according to any one of claims 1 to 12 are implemented.

14. An image recognition system, characterized in that: The recognition system includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the steps of the image recognition method for strip-shaped articles according to any one of claims 1 to 12 are implemented.

15. A refrigeration device, characterized in that: The refrigeration equipment includes an image recognition system, which includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor executes the computer program, the steps of the image recognition method for strip-shaped articles described in any one of claims 1 to 12 are implemented.

Images