Visual recognition system and method based on artificial intelligence

Abstract

The present invention belongs to the field of computer vision technology. A visual recognition system and method based on artificial intelligence are provided. The method comprises: controlling a camera to capture several frames of high-definition images of a first target product and a second target product adjacent to each other on a conveyor belt; extracting a first dynamic feature of the first target product and a second dynamic feature of the second target product from respective high-definition images; inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the probability of qualified packaging of the first target product and the second target product; and marking a qualified packaging label or a unqualified packaging label for the first target product and the second target product according to the qualified packaging probability. The technical solution provided by the present invention solves the problem that the image quality inspection of whether the packaging is qualified cannot be performed due to the mutual compression of adjacent products, reduces the missed detection rate of unqualified packaging, and is conducive to improving product quality.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a visual recognition system and method based on artificial intelligence. Background Art

[0002] Using image recognition technology to inspect the outer packaging of packaged products is an important part of production quality inspection. This quality inspection mainly involves whether the outer packaging is intact. For example, if the cover of the packaging box is not fastened properly, it will be judged as unqualified. However, due to the shooting angle of the quality inspection camera, it is difficult to capture the bottom of the product packaging box on the conveyor belt, resulting in the inability to effectively inspect the bottom of the product packaging box. For example, Figure 1 As shown, there are adjacent first target products and second target products on the conveyor belt, wherein the packaging of the first target product is unqualified, and its cover is not fastened, and is covered by the second target product that falls on the conveyor belt later. Since the first target product and the second target product are too close to each other, the camera cannot capture the unfastened cover, which makes it easy to misjudge the first target product as a packaged qualified product.

[0003] This technical problem has not yet been effectively solved, resulting in a high missed inspection rate in the quality inspection of product outer packaging, which cannot meet actual production needs. Summary of the invention

[0004] In this regard, the present invention provides a visual recognition method, system, electronic device, computer storage medium and computer program product based on artificial intelligence to solve the above-mentioned technical problems.

[0005] The present invention discloses a visual recognition method based on artificial intelligence, which comprises the following steps:

[0006] Controlling a camera to capture a plurality of first high-definition images of a first target product on a conveyor belt, and a plurality of second high-definition images of a second target product; wherein the first target product and the second target product are adjacent to each other on the conveyor belt and run toward the camera, and the first target product is located in front of the second target product;

[0007] Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image;

[0008] Inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product;

[0009] The first target product and the second target product are marked with a qualified packaging label or a unqualified packaging label according to the qualified packaging probability.

[0010] In some embodiments, extracting a first dynamic feature of the first target product from the first high-definition image and extracting a second dynamic feature of the second target product from the second high-definition image include:

[0011] Extracting first feature point position information of the front end and second feature point position information of the rear end of the first target product from each frame of the first high-definition image, associating the first feature point position information with the second feature point position information, and determining each group of the associated first feature point position information and the second feature point position information as the first dynamic feature of the first target product;

[0012] The third feature point position information of the front end and the fourth feature point position information of the rear end of the second target product are extracted from each frame of the second high-definition image, the third feature point position information and the fourth feature point position information are associated, and each group of the associated third feature point position information and the fourth feature point position information are determined as the second dynamic feature of the second target product.

[0013] In some embodiments, before controlling the camera to capture a plurality of first high-definition images of the first target product on the conveyor belt and a plurality of second high-definition images of the second target product, the method further includes:

[0014] Control the camera to capture a third high-definition image of each product to be inspected on the conveyor belt, and extract the position data and adjacent distance data of each product to be inspected based on the third high-definition image;

[0015] If the position and posture data meets the first condition and the adjacent distance data meets the second condition, the product to be inspected is determined as the second target product, and the product to be inspected adjacent to the product to be inspected is determined as the first target product;

[0016] The first condition is specifically: the difference between a first height of the front end of the product to be inspected close to the camera and a second height of the rear end away from the camera is greater than a first threshold;

[0017] The second condition is specifically: the distance between the product to be inspected and its adjacent product to be inspected in front is less than a second threshold.

[0018] In some embodiments, the first threshold is determined by:

[0019] Acquire the three-dimensional packaging design data of the products to be inspected in this batch, determine the distance difference between the front end and the rear end according to the three-dimensional packaging design data, and determine the first threshold value based on the distance difference and a first adjustment coefficient; wherein the first adjustment coefficient is greater than 1;

[0020] And, the second threshold is determined by:

[0021] A plurality of design lengths related to the cover plate are determined based on the three-dimensional packaging design data, and a maximum value of the design lengths is determined as the second threshold.

[0022] In some embodiments, the first dynamic feature and the second dynamic feature are input into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product, including:

[0023] Sort the first feature point position information and the second feature point position information in the first dynamic feature in chronological order;

[0024] Calculating first differences between the position information of the first feature points that are adjacent after sorting, and constructing a first fluctuation data set according to the first differences; and calculating second differences between the position information of the second feature points that are adjacent after sorting, and constructing a second fluctuation data set according to the second differences; performing time registration on the first fluctuation data set and the second fluctuation data set to obtain a third fluctuation data set of the first target product;

[0025] Calculate and obtain a fourth fluctuation data set of the second target product in the same manner;

[0026] The third fluctuation data set and the fourth fluctuation data set are simultaneously input into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product.

[0027] In some embodiments, the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product, including:

[0028] The quality inspection prediction model outputs the preliminary packaging qualified probability of the first target product and the second target product;

[0029] The structural strength of the cover plate is determined according to the three-dimensional packaging design data, a second adjustment coefficient is determined according to the structural strength, and the preliminary packaging qualified probability is corrected to the packaging qualified probability using the second adjustment coefficient.

[0030] The present invention also discloses a visual recognition system based on artificial intelligence, the system comprising a control module, a storage module, and a camera; the control module implements the following method steps by calling and executing the computer code in the storage module:

[0031] Controlling a camera to capture a plurality of first high-definition images of a first target product on a conveyor belt, and a plurality of second high-definition images of a second target product; wherein the first target product and the second target product are adjacent to each other on the conveyor belt and run toward the camera, and the first target product is located in front of the second target product;

[0032] Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image;

[0033] Inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product;

[0034] The first target product and the second target product are marked with a qualified packaging label or a unqualified packaging label according to the qualified packaging probability.

[0035] The present invention also discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in any of the preceding items.

[0036] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0037] The present invention also discloses a computer program product. When the computer program product is run on a terminal, the terminal implements any of the above methods when executing the computer program product.

[0038] The beneficial effect of the present invention is that the technical solution provided by the present invention solves the problem of being unable to perform image quality inspection on whether the packaging is qualified due to the mutual compression of adjacent products, reduces the missed detection rate of unqualified packaging, and is conducive to improving product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 is a schematic diagram of a scenario of quality inspection of outer packaging of a product based on image recognition technology disclosed in an embodiment of the present invention;

[0041] Figure 2 It is a flowchart of a visual recognition method based on artificial intelligence disclosed in an embodiment of the present invention;

[0042] Figure 3 It is a structural schematic diagram of a visual recognition system based on artificial intelligence disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0044] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0045] As described in the background technology, the product packaging to be inspected on the conveyor belt has a top cover and a bottom cover that are not fastened properly. The unfastened top cover and / or bottom cover may be easily pressed underneath by the closely adjacent qualified packaged products, causing the camera to be unable to capture the unfastened top cover and / or bottom cover. As a result, the unqualified packaged products may be misjudged as qualified packaged products, that is, the missed inspection rate of unqualified packaged products is too high.

[0046] In view of the above technical problems, the embodiment of the present invention discloses a visual recognition method based on artificial intelligence, such as Figure 1 , Figure 2 As shown, the method comprises the following steps:

[0047] Controlling a camera to capture a plurality of first high-definition images of a first target product on a conveyor belt, and a plurality of second high-definition images of a second target product; wherein the first target product and the second target product are adjacent to each other on the conveyor belt and run toward the camera, and the first target product is located in front of the second target product;

[0048] Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image;

[0049] Inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product;

[0050] The first target product and the second target product are marked with a qualified packaging label or a unqualified packaging label according to the qualified packaging probability.

[0051] The present invention uses high-definition images captured by a camera to perform quality inspection on the outer packaging of products transmitted on a conveyor belt. Specifically, the camera is controlled to capture several frames of high-definition images of mutually adjacent target products on the conveyor belt, and the first dynamic features and the second dynamic features of the two are respectively extracted from the high-definition images. Then, the extracted first dynamic features and the second dynamic features are input into a quality inspection prediction model pre-built based on artificial intelligence technology, and the quality inspection prediction model can output the qualified packaging probability of the two products. Finally, based on the qualified packaging probability of the two, a qualified packaging label or an unqualified packaging label is marked for the first target product or the second target product. For example, when the qualified packaging probability is higher than the probability threshold, a qualified packaging label is marked accordingly, otherwise a unqualified packaging label is marked accordingly. Subsequently, relevant personnel can be reminded to conduct manual quality inspection based on the label.

[0052] Based on the above technical solution, the present invention solves the problem of being unable to perform image quality inspection on whether the packaging is qualified due to the mutual compression of adjacent products, reduces the missed inspection rate of unqualified packaging, and is conducive to improving product quality.

[0053] In some embodiments, extracting a first dynamic feature of the first target product from the first high-definition image and extracting a second dynamic feature of the second target product from the second high-definition image include:

[0054] Extracting first feature point position information of the front end and second feature point position information of the rear end of the first target product from each frame of the first high-definition image, associating the first feature point position information with the second feature point position information, and determining each group of the associated first feature point position information and the second feature point position information as the first dynamic feature of the first target product;

[0055] The third feature point position information of the front end and the fourth feature point position information of the rear end of the second target product are extracted from each frame of the second high-definition image, the third feature point position information and the fourth feature point position information are associated, and each group of the associated third feature point position information and the fourth feature point position information are determined as the second dynamic feature of the second target product.

[0056] In the embodiment of the present invention, the present invention analyzes whether there is a situation where the cover plate is not fastened properly and is pressed by analyzing the dynamic characteristics of the adjacent first target product and the second target product during the transmission process on the conveyor belt. Based on this technical principle, the present invention extracts the position information of the designated feature points (such as vertices, which can be multiple) of the front and rear ends of the target product from the multiple frames of high-definition images taken by the camera, and associates the feature point position information extracted from each frame of high-definition image, and then determines the obtained associated data group as the dynamic characteristics of the corresponding target product. Subsequently, the quality inspection prediction model simultaneously conducts an in-depth analysis of the dynamic characteristics of the adjacent target products, and can analyze whether the two are in the situation where the cover plate is not fastened properly and is pressed.

[0057] In some embodiments, before controlling the camera to capture a plurality of first high-definition images of the first target product on the conveyor belt and a plurality of second high-definition images of the second target product, the method further includes:

[0058] Control the camera to capture a third high-definition image of each product to be inspected on the conveyor belt, and extract the position data and adjacent distance data of each product to be inspected based on the third high-definition image;

[0059] If the position and posture data meets the first condition and the adjacent distance data meets the second condition, the product to be inspected is determined as the second target product, and the product to be inspected adjacent to the product to be inspected is determined as the first target product;

[0060] The first condition is specifically: a difference between a first height of the front end of the product to be inspected close to the camera and a second height of the rear end away from the camera is greater than a first threshold;

[0061] The second condition is specifically: the distance between the product to be inspected and its adjacent product to be inspected in front is less than a second threshold.

[0062] In an embodiment of the present invention, only when the distance between the adjacent products to be inspected on the conveyor belt is too small will the camera fail to capture the unfastened cover of the first target product in front. At the same time, after the unfastened cover of the front product is pressed and covered by the rear product, the posture of the rear product will be high in the front and low in the back. Therefore, based on the distance between the product to be inspected and the adjacent product in front and the posture of the product to be inspected itself, the situation where the unfastened cover is pressed and covered by the rear product is screened to quickly determine the first target product and the second target product for targeted analysis. In this way, the number of in-depth analyses of the above-mentioned pressing situation can be effectively reduced to reduce the amount of data processing and improve the quality inspection efficiency. In addition, for those products to be inspected that do not meet the first and second conditions, the above-mentioned pressing situation can be basically ruled out, and conventional image quality inspection methods can be used, which is conducive to improving the overall quality inspection efficiency.

[0063] Specifically, the camera is first controlled to capture the third high-definition image (which can be a video or a multi-frame static image) of each product to be inspected on the conveyor belt, and the posture data and adjacent distance data of each product to be inspected are extracted from it. The posture data refers to the difference between the first height of the product to be inspected being analyzed close to the front end of the camera and the second height away from the rear end of the camera; the adjacent distance data refers to the distance between the product to be inspected being analyzed and the adjacent product to be inspected in front of it.

[0064] If the difference between the above-mentioned first height and the second height is greater than the first threshold value, it means that the front end of the product to be inspected being analyzed is "tilted", indicating that the above-mentioned capping situation may exist, that is, the unfastened cover of the front product will be lifted up at the front end of the rear product, thereby presenting the above-mentioned front end "tilted" situation. When the above-mentioned capping situation does not exist, the top of the rear product is in a "horizontal" state, and the difference between the first height data and the second height data will not be higher than the first threshold value. In addition, since the length of the cover of the product to be inspected is limited, the above-mentioned capping situation will only occur when the distance between the front and rear adjacent products to be inspected is too close. Therefore, the present invention is arranged that only when the above-mentioned first condition and the second condition are met, the product to be inspected being analyzed is determined as the second target product, and the product to be inspected adjacent to it is determined as the first target product.

[0065] In some embodiments, the first threshold is determined by:

[0066] Acquire the three-dimensional packaging design data of the products to be inspected in this batch, determine the distance difference between the front end and the rear end according to the three-dimensional packaging design data, and determine the first threshold value based on the distance difference and a first adjustment coefficient; wherein the first adjustment coefficient is greater than 1;

[0067] And, the second threshold is determined by:

[0068] A plurality of design lengths related to the cover plate are determined based on the three-dimensional packaging design data, and a maximum value of the design lengths is determined as the second threshold.

[0069] In the embodiment of the present invention, the outer packaging structures of products in different batches are different. The distance difference between the front and rear ends of the products to be inspected in this batch can be determined through the three-dimensional packaging design data. At the same time, a first adjustment coefficient is determined based on several factors. The first adjustment coefficient is used to appropriately increase the distance difference, that is, to obtain the above-mentioned first threshold. Therefore, when the height difference between the front and rear ends of the product to be inspected is greater than the first threshold, it is determined that it has a necessary condition that the cover plate of the previous product is not fastened properly.

[0070] At the same time, the cover plate in the three-dimensional packaging design data can also be identified. The cover plate may include multiple folding plates of different lengths, such as an upper cover plate and a lower cover plate. The maximum value of the design lengths corresponding to each of these cover plates is determined as the second threshold mentioned above. Therefore, when the distance between the adjacent products to be inspected is less than the second threshold, it is determined that they have another necessary condition of covering the previous product that has not been fastened properly.

[0071] In some embodiments, the first dynamic feature and the second dynamic feature are input into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product, including:

[0072] Sort the first feature point position information and the second feature point position information in the first dynamic feature in chronological order;

[0073] Calculating first differences between the position information of the first feature points that are adjacent after sorting, and constructing a first fluctuation data set according to the first differences; and calculating second differences between the position information of the second feature points that are adjacent after sorting, and constructing a second fluctuation data set according to the second differences; performing time registration on the first fluctuation data set and the second fluctuation data set to obtain a third fluctuation data set of the first target product;

[0074] Calculate and obtain a fourth fluctuation data set of the second target product in the same manner;

[0075] The third fluctuation data set and the fourth fluctuation data set are simultaneously input into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product.

[0076] In an embodiment of the present invention, the first dynamic feature and the second dynamic feature obtained above respectively include the feature point change data of the front and rear ends of the first target product and the second target product, such as the first feature point position information of the front end and the second feature point position information of the rear end of the first target product. These feature point position information of the front end of the first target product are sorted in chronological order according to the acquisition time, and then the difference between the adjacent feature point position information (i.e., the position change distance) is calculated. These differences constitute the first fluctuation data set; similarly, the second fluctuation data set of the rear end can be calculated. The first fluctuation data set of the front end and the second fluctuation data set of the rear end are time-aligned to obtain the third fluctuation data set. The aligned third fluctuation data set includes multiple sets of up and down fluctuation data of the front end and the rear end of the first target product corresponding to the same time.

[0077] Similarly, the fourth fluctuation data set (corresponding to the third fluctuation data set) of the second target product can be calculated. The third fluctuation data set of the first target product and the fourth fluctuation data set of the second target product are simultaneously input into the quality inspection prediction model. The quality inspection prediction model can simultaneously conduct an in-depth analysis of the up and down fluctuation conditions of the front and rear ends of the first target product and the second target product. When the up and down fluctuation conditions of the front and rear ends of the first target product and the second target product meet the above-mentioned capping conditions, the probability of packaging qualification can be determined based on the respective compliance conditions of the first target product and the second target product.

[0078] It should be noted that when the above-mentioned capping situation exists, the front end of the second target product is supported by the cover plate below, and its front end has a higher up and down fluctuation amplitude and frequency than the rear end, while the rear end of the first target product in front will have a higher up and down fluctuation amplitude and frequency than its front end due to the tensile stress of the front end of the second target product behind (but the amplitude will be lower than that of the second target product). By deeply training the quality inspection prediction model with labeled training data, it can grasp the up and down fluctuation characteristics of the front and rear ends of the products to be inspected before and after the capping situation exists, and then realize the accurate identification of the capping situation, and output the qualified probability of the front and rear adjacent packaging to be inspected with poor evaluation.

[0079] In some embodiments, the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product, including:

[0080] The quality inspection prediction model outputs the preliminary packaging qualified probability of the first target product and the second target product;

[0081] The structural strength of the cover plate is determined according to the three-dimensional packaging design data, a second adjustment coefficient is determined according to the structural strength, and the preliminary packaging qualified probability is corrected to the packaging qualified probability using the second adjustment coefficient.

[0082] In the embodiment of the present invention, due to the different materials, thicknesses, etc. used in the cover plate, there are large differences in its structural strength, and different structural strengths will have different effects on the up and down fluctuation amplitude of the front end of the second target product that is pressed. Specifically, when the structural strength of the cover plate is higher, the impact on the up and down fluctuation amplitude of the front end of the second target product that is pressed is smaller, that is, in this case, the up and down fluctuation amplitude of the front end of the second target product is mainly caused by the own fluctuation of the conveyor belt; and when the structural strength of the cover plate is lower, it is softer, and it is difficult to provide stable support for the front end of the second target product that is pressed, resulting in a larger up and down fluctuation amplitude of the front end of the second target product, that is, the greater the impact on the up and down fluctuation amplitude of the front end of the second target product that is pressed. Correspondingly, the tensile stress on the rear end of the first target product when the second target product fluctuates up and down is also the same. Of course, the above-mentioned second adjustment coefficient can also be used only for the correction of the initial packaging qualified probability of the second target product, and not for the correction of the initial packaging qualified probability of the first target product, and the present invention is not specifically limited to this.

[0083] In this regard, after the quality inspection prediction model outputs the preliminary packaging qualified probability of the first target product and the second target product, the present invention determines the second adjustment coefficient according to the structural strength of the cover plate, and uses the second adjustment coefficient to correct the preliminary packaging qualified probability to the packaging qualified probability. The structural strength of the cover plate can be extracted from the aforementioned three-dimensional packaging design data, such as the material and thickness of the cover plate included in the three-dimensional packaging design data.

[0084] like Figure 3 As shown, the embodiment of the present invention also discloses a visual recognition system based on artificial intelligence, the system includes a control module, a storage module, and a camera; the control module implements the following method steps by calling and executing the computer code in the storage module:

[0085] Controlling a camera to capture a plurality of first high-definition images of a first target product on a conveyor belt, and a plurality of second high-definition images of a second target product; wherein the first target product and the second target product are adjacent to each other on the conveyor belt and run toward the camera, and the first target product is located in front of the second target product;

[0086] Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image;

[0087] Inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product;

[0088] The first target product and the second target product are marked with a qualified packaging label or a unqualified packaging label according to the qualified packaging probability.

[0089] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in the above embodiment.

[0090] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method described in the above embodiment.

[0091] The embodiment of the present invention further discloses a computer program product. When the computer program product is run on a terminal, the terminal implements the method described in the above embodiment.

[0092] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0093] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0094] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A visual recognition method based on artificial intelligence, characterized in that ,include: Controlling a camera to capture a plurality of first high-definition images of a first target product on a conveyor belt, and a plurality of second high-definition images of a second target product; wherein the first target product and the second target product are adjacent to each other on the conveyor belt and run toward the camera, and the first target product is located in front of the second target product; Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image; the first dynamic feature and the second dynamic feature respectively include position change information of a plurality of feature points at the front and rear ends of the first target product and the second target product; Inputting the first dynamic feature and the second dynamic feature into a quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputting the packaging qualified probability of the first target product and the second target product, including: The feature point position information of the front end of the first target product is sorted in the order of acquisition time, and then the differences between the adjacent feature point position information are calculated, and these differences constitute the first fluctuation data set; similarly, the second fluctuation data set of the back end is calculated; the first fluctuation data set of the front end and the second fluctuation data set of the back end are time-aligned, that is, the third fluctuation data set of the first target product is obtained, and the fourth fluctuation data set of the second target product is calculated in the same way; the third fluctuation data set and the fourth fluctuation data set are simultaneously input into the quality inspection prediction model based on artificial intelligence, and the quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product; The first target product and the second target product are marked with a qualified packaging label or a unqualified packaging label according to the qualified packaging probability.

2. The method of visual recognition based on artificial intelligence according to claim 1, characterized in that: Extracting a first dynamic feature of the first target product from the first high-definition image, and extracting a second dynamic feature of the second target product from the second high-definition image, comprises: Extracting first feature point position information of the front end and second feature point position information of the rear end of the first target product from each frame of the first high-definition image, associating the first feature point position information with the second feature point position information, and determining each group of the associated first feature point position information and the second feature point position information as the first dynamic feature of the first target product; The third feature point position information of the front end and the fourth feature point position information of the rear end of the second target product are extracted from each frame of the second high-definition image, the third feature point position information and the fourth feature point position information are associated, and each group of the associated third feature point position information and the fourth feature point position information are determined as the second dynamic feature of the second target product.

3. The method of visual recognition based on artificial intelligence according to claim 2, characterized in that: Before controlling the camera to capture a plurality of first high-definition images of the first target product on the conveyor belt and a plurality of second high-definition images of the second target product, the method further includes: Control the camera to capture a third high-definition image of each product to be inspected on the conveyor belt, and extract the position data and adjacent distance data of each product to be inspected based on the third high-definition image; If the position and posture data meets the first condition and the adjacent distance data meets the second condition, the product to be inspected is determined as the second target product, and the product to be inspected adjacent to the product to be inspected is determined as the first target product; The first condition is specifically: a difference between a first height of the front end of the product to be inspected close to the camera and a second height of the rear end away from the camera is greater than a first threshold; The second condition is specifically: the distance between the product to be inspected and its adjacent product to be inspected in front is less than a second threshold.

4. The method for visual recognition based on artificial intelligence according to claim 3, characterized in that: The first threshold is determined by: Acquire the three-dimensional packaging design data of the products to be inspected in this batch, determine the distance difference between the front end and the rear end according to the three-dimensional packaging design data, and determine the first threshold value based on the distance difference and a first adjustment coefficient; wherein the first adjustment coefficient is greater than 1; And, the second threshold is determined by: A plurality of design lengths related to the cover plate are determined based on the three-dimensional packaging design data, and a maximum value of the design lengths is determined as the second threshold.

5. The method of visual recognition based on artificial intelligence according to claim 4, characterized in that: The quality inspection prediction model outputs the packaging qualified probability of the first target product and the second target product, including: The quality inspection prediction model outputs the preliminary packaging qualified probability of the first target product and the second target product; The structural strength of the cover plate is determined according to the three-dimensional packaging design data, a second adjustment coefficient is determined according to the structural strength, and the preliminary packaging qualified probability is corrected to the packaging qualified probability using the second adjustment coefficient.

6. A visual recognition system based on artificial intelligence, the system comprising a control module, a storage module, and a camera; characterized in that: The control module implements the method according to any one of claims 1 to 5 by calling and executing the computer code in the storage module.

7. An electronic device comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 5.

8. A computer storage medium storing a computer program, characterized in that: The computer program is executed by a processor to implement the method according to any one of claims 1 to 5.

9. A computer program product comprising a computer program stored on a non-transitory computer readable medium, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images