Production line product appearance detection method and system

Abstract

The application relates to a production line product appearance detection method and system and a computer readable storage medium. According to the method, the following steps are included: acquiring one or more sample photos of one or more to-be-detected parts of a product on a detection station and label information of the product; labeling the one or more sample photos based on the label information of the product; using the labeled one or more sample photos as training samples to perform appearance detection model training for a type of product; and detecting the appearance of a subsequent product of the same type on a production line by using the appearance detection model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of detection, in particular, to a production line product appearance detection method and system, and a computer readable storage medium. BACKGROUND

[0002] In the production of complex production line products, there are many types and quantities of products on the same production line, and these products need to be detected one by one at the detection station. However, the detection defects and standards of each part of the product are not always consistent for different types of products, and special detection needs to be performed for specific parts of specific products. For example, on the production line of an automobile, there can be automobile products of multiple vehicle models, and each vehicle model has multiple automobile products. These different vehicle models of automobiles are scattered in various workstations on the production line (such as installation and assembly) for production work, and different vehicle models of automobile products are mixed and sent to the detection station.

[0003] In the current product appearance detection of machine learning known to the inventors, a photograph needs to be taken for each part to be detected of a certain type of product to establish a training sample library for that part to be detected of that type of product. After training the appearance detection model using the training samples, the appearance detection model can be debugged and used on the detection equipment to detect the appearance of the product. Therefore, the labeling of each training sample will be very complicated, time-consuming and labor-intensive, and cannot be automatically connected to the detection of the appearance of the product after the model training is completed. Different products need to be trained separately, and even for the same product, after the same product is updated in the future, retraining is required. SUMMARY

[0004] In view of the above problems, the present application aims to improve the training and detection efficiency of production line products by providing a production line product appearance detection method and system, and a computer readable storage medium.

[0005] According to a first aspect of the present application, a production line product appearance detection method is provided, characterized in that the production line includes a detection station, and the production line has one or more types of products, each type having one or more products, and each type of product has one or more parts to be detected. The method comprises the following steps: obtaining one or more sample photos of one or more parts to be detected of the product on the detection station and the label information of the product; labeling the one or more sample photos based on the label information of the product; using the labeled one or more sample photos as training samples to train an appearance detection model for the type of product; and using the appearance detection model to detect the appearance of the subsequent product of the same type on the production line.

[0006] The method according to an embodiment of the present application, wherein the method further comprises iteratively training the appearance detection model using the detected sample photos of the products.

[0007] The method according to an embodiment of the present application, wherein the method further comprises, in response to a sample photo of a subsequent product being detected as unqualified, transmitting the sample photo of the product detected as unqualified to a manual judgment device for judgment.

[0008] The method according to an embodiment of the present application, wherein the labeling further comprises establishing a classification table for each to-be-detected part of the product, so as to classify sample photos of the same type of product, for training of the appearance detection model.

[0009] The method according to an embodiment of the present application, wherein the training of the appearance detection model further comprises, in response to different to-be-detected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-detected part of each type of product; and in response to different to-be-detected parts of different types of products in one or more types of products on the production line having the same sample photos, merging the classification of the same sample photos.

[0010] The method according to an embodiment of the present application, wherein the training of the appearance detection model further comprises, in response to different to-be-detected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-detected part of each type of product; and in response to different to-be-detected parts of different types of products in one or more types of products on the production line having the same sample photos, transmitting the same sample photos to a manual judgment device for judgment, and in response to the judgment passing, merging the classification of the same sample photos.

[0011] According to a second aspect of the present application, a production line product appearance detection system is provided, the production line comprising a detection station, the production line having one or more types of products, each type having one or more products, and each type of product having one or more to-be-detected parts, the system comprising: an acquisition device configured to acquire one or more sample photos of one or more to-be-detected parts of a product on the detection station and label information of the product; a memory storing instructions; and a processor configured to execute the instructions to perform the following operations:

[0012] labeling the one or more sample photos based on the label information of the product;

[0013] training an appearance detection model for the type of product using the labeled one or more sample photos as training samples; and

[0014] detecting, by the appearance detection model, the appearance of a subsequent product of the same type on the production line.

[0015] According to an embodiment of the system of the present application, the processor is further configured to execute instructions to perform iterative training of the appearance detection model using the sample photos of the detected products.

[0016] According to an embodiment of the system of the present application, the processor is further configured to execute instructions to perform, in response to the sample photo of the subsequent product being detected as unqualified, transmitting the sample photo corresponding to the detected unqualified product to the manual judgment device for judgment.

[0017] According to an embodiment of the system of the present application, the processor is further configured to execute instructions to perform establishing a classification table for each to-be-detected part of the product to classify the sample photos of the products of the same type for training of the appearance detection model.

[0018] According to an embodiment of the system of the present application, the processor is further configured to execute instructions to perform, in response to different to-be-detected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-detected part of each product type; and in response to different to-be-detected parts of different types of products in one or more types of products on the production line having the same sample photos, merging the classifications of the same sample photos.

[0019] According to an embodiment of the system of the present application, the processor is further configured to execute instructions to perform, in response to different to-be-detected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-detected part of each product type; and in response to different to-be-detected parts of different types of products in one or more types of products on the production line having the same sample photos, transmitting the same sample photos to the manual judgment device for judgment, and in response to the judgment passing, merging the classifications of the same sample photos.

[0020] According to a third aspect of the present application, a computer readable storage medium is provided, the computer readable storage medium storing instructions which, when executed by a processor, perform the production line product appearance detection method according to any one of the preceding embodiments.

[0021] The present application greatly reduces the time and effort cost of manual annotation through the detection sample acquisition and annotation automation of various types of products on the production line. The acquisition process of the detection sample, the training process and the appearance detection process of the product can be automatically connected, greatly improving the efficiency of the production line product detection. In addition, the detection sample of the detected product can be automatically input into the model as a new training sample for iterative training, further improving the accuracy of the model detection. BRIEF DESCRIPTION OF DRAWINGS

[0022] The above and other objects and advantages of the present application will become more apparent by describing in detail the following non-restrictive embodiments thereof in conjunction with the accompanying drawings, in which like reference numerals refer to like elements throughout.

[0023] Figure 1 A flowchart of a production line product appearance detection method 100 according to some embodiments is shown.

[0024] Figure 2 A module diagram of a production line product appearance detection system 200 according to some embodiments is shown. DETAILED DESCRIPTION

[0025] The following detailed description of implementations is merely exemplary in nature and is not intended to limit or otherwise restrict the disclosed technology and applications and uses thereof. Furthermore, there is no intention to be bound by any expressed or implied theory presented in the preceding technical field, background or the following detailed description.

[0026] In the following detailed description of embodiments, numerous specific details are set forth in order to provide a more thorough understanding of the disclosed technology. However, it will be apparent to one of ordinary skill in the art having the benefit of this disclosure that the technology disclosed can be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.

[0027] The terms such as "comprising" and "including" mean that the technical solution of the present application does not exclude the presence of other elements or steps not directly or explicitly stated in the specification.

[0028] In some embodiments of the present application, the implementation of appearance detection of products on a production line can roughly include the following process: in the production of products on a production line, the same production line can involve the production process of various products, such as the installation, addition, etc. of products. For each product, a photo of each part to be detected of the product can be obtained on a detection station, such as taking a photo of the product by using one or more cameras arranged at different shooting angles to obtain one or more photos including a certain part to be detected. In this process, the equipment on the detection station also obtains the label information of the product (such as model, date, etc. identification information), and labels and records the photos with the label information of the product, which can make each photo have the type of the product indicated by the photo and other label information. For each part to be detected of each type of product, a library including multiple photos of the part to be detected can be obtained, and the photos in the library can be used as training samples to train an appearance detection model for the part to be detected of the type. When the next product of the type reaches the detection station on the production line, the appearance detection model for the part to be detected of the type can be used for detection, thereby realizing the automatic connection of the detection training model process and the detection process on the production line.

[0029] The inventors of the present application found that there are often multiple types of products being installed on a production line at the same time, so the products passing through the detection station can be constantly changing, in addition, even for the same part to be detected of the same type of product, the detection standard can change, and the product itself can also be updated and upgraded. However, the current product detection combined with machine learning is static and separate, that is, multiple products of a certain type need to be taken off the production line for special training sample acquisition and labeling (i.e. taking photos of each part to be detected and labeling each photo), after obtaining some training samples, a computing device is used to train an appearance detection model, after the training is completed, the appearance detection model for the individual type of product needs to be debugged to the equipment on the detection station, and then the equipment on the detection station can start to use the trained appearance detection model for product detection. Therefore, if the production line has multiple types of products, multiple products of each type of product need to be taken off line for additional training sample acquisition and model training, whether manually or by using special equipment, the workload of acquiring photos and labeling is very large, and the four processes of acquisition, labeling, training and detection cannot be automatically and smoothly connected on the flowing production line, and the production line needs to be interrupted, which reduces the production efficiency and the detection cost is also very high.

[0030] The above-mentioned embodiments realize the automatic implementation of the four processes of acquisition, labeling, training and detection under the premise of uninterrupted production line, improve the production efficiency, and greatly reduce the huge workload of labeling and training.

[0031] It should be noted that the defects detected herein include any defects that can be embodied on the appearance of the product, which can be suitable for being achieved via image detection. For example, deformation of the appearance of the product, uniformity of coloring, color difference, scratch, breakage, missing parts, misassembled parts, etc.

[0032] Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings.

[0033] Figure 1 A flowchart of a production line product appearance detection method 100 according to some embodiments is shown. There are detection stations on the production line, and there are one or more types of products on the production line, each type having multiple products, and for each type of product, there are one or more parts to be detected. The appearance detection method 100 includes the following steps:

[0034] In step 110, one or more sample photos of one or more parts to be detected of the product on the detection station and the label information of the product are obtained.

[0035] In some examples, the photos of the product can be obtained by using the cameras installed on the detection station, which can be multiple cameras arranged to take photos of the product from various angles to obtain photos of each part to be detected of the product. The cameras can also recognize the label of the product, such as the label sticker, two-dimensional code, etc. of the product, and perform label recognition, or the cameras transmit the photos with label information obtained by shooting to the processing device, and the recognition is performed in the processing device. The processing device here can be the same device as the processing device for detection or training in the subsequent steps, or it can be another processing device. Of course, the recognition of the label can also be performed by using a special device, such as a scanner for two-dimensional code, etc. Thus, for each part to be detected of each type of photo, one or more photos including the part to be detected are obtained, and the label information (such as model number, production date, production batch, version number, etc.) of the product is also obtained.

[0036] In some examples, the obtained photos can also be automatically cropped, and most of the content of the cropped photos is for a certain part to be detected. In this way, the quality and information richness of the photos for subsequent training as training samples can be further improved, so that the subsequent training and detection process is more accurate and effective.

[0037] In step 120, one or more sample photos are labeled based on the label information of the product.

[0038] This step can be automatically performed after step 110, so that each sample photo has the product label information corresponding to the sample photo, and thus it can be known which part of which type of product the sample photo is from. This is beneficial for the classification of each part to be detected of each type of product, so as to establish a separate training sample library, so that the appearance detection model training for each specific part to be detected is more specialized, and the training effect and detection accuracy are improved. For example, typically, a standard library (table) can be established for all types of parts to be detected of all types of products, which is distinguished by product type, for example, and is distinguished by different parts to be detected in the same type of product. For example, there are x type and y type vehicles on a certain automobile production line, and the parts to be detected of the x type vehicle may, for example, include wheels, lights, and windows, and the parts to be detected of the y type vehicle may, for example, also include wheels, lights, and windows, and then six classifications (for example, six folders are established) can be included: x type wheels, x type lights, x type windows, y type wheels, y type lights, and y type windows, and each classification includes one or more photos of the part.

[0039] In step 130, the appearance detection model for the type of product is trained using the labeled one or more sample photos as training samples.

[0040] As described above, each part to be detected of each type of product has a certain training sample photo, and each part to be detected can be trained according to the classification of the standard library (table) described above. For example, there are x type and y type vehicles on a certain automobile production line, and the parts to be detected of the x type vehicle may, for example, include wheels, lights, and windows, and the parts to be detected of the y type vehicle may, for example, also include wheels, lights, and windows, and then six appearance detection models are trained for the six parts to be detected of the two types, i.e., an appearance detection model for the wheels of the x type, an appearance detection model for the lights of the x type, an appearance detection model for the windows of the x type, an appearance detection model for the wheels of the y type, an appearance detection model for the lights of the y type, and an appearance detection model for the windows of the y type.

[0041] In step 140, the appearance of the subsequent products of the same type on the production line is detected using the appearance detection model.

[0042] After the appearance training model of each part of a certain type of product is obtained, the equipment on the detection station can automatically perform appearance detection on the next product of the same type of work being detected to identify whether the product has defects in each part. If the next product to arrive at the detection station is another type of product, the above steps 110-140 can be started to perform the process on the other type of product. The detection result can be displayed, for example, as a photo of a certain part is unqualified / qualified, and the worker at the detection station is prompted in a certain way (such as display of a human-machine interface, alarm, red-green light flashing, etc.). Alternatively, a matching score is determined, for example, the detection matching score of a certain part to be detected can reflect the matching degree between the sample photo of the part to be detected and the trained and qualified photo in the appearance detection model, and a certain matching degree (such as 80% matching) is used as the basis for whether it is qualified.

[0043] Therefore, the method 100 of the present application can respectively establish a training sample and an appearance detection model for each part to be detected of each type of product, and then detect subsequent products of each type, so that the product production process, model training process, product detection process, etc. on the production line can be carried out at the same time, greatly improving the production efficiency and reducing the workload of training sample preparation work (such as labeling, cropping, manual input, etc.).

[0044] In some embodiments, the method 100 described above can further include an iterative training step, i.e., after the appearance detection for a certain product is completed, the appearance detection model can be iteratively trained using the detected sample photos of the product. For example, taking the x and y models of the vehicle in the above example of the vehicle production line as an example, after the training sample acquisition and training for a certain x model of the vehicle are completed, the next vehicle of the x model can be detected, and the detection process of the next vehicle can be, for example, comparing the photos of each part to be detected of the vehicle with the trained samples in the corresponding trained appearance detection model, and determining according to certain comparison rules, such as machine learning / deep neural network comparison and detection. The result of the determination can be, for example, unqualified / qualified, or have the matching score described above. Then, the sample photos determined to be qualified can be used as further training samples to positively train the appearance detection model of the part to be detected corresponding to the photos, and the sample photos determined to be unqualified can be used as further training samples to negatively train the appearance detection model of the part to be detected corresponding to the photos, so as to further improve the training effect and detection effect of the appearance detection model. Alternatively, the matching score described above can also be used as a weight coefficient for subsequent training of the appearance detection model, for example, the higher the matching score of the sample, the higher the training weight of the sample as a new training sample in the neural network / deep learning. It can be understood that as the number of detections increases, the optimization degree of the appearance detection model is higher and higher.

[0045] In some embodiments, the method 100 described above can further include a step of combining human judgment, i.e., in response to the sample photos of the subsequent products of a certain type of product being detected as unqualified, the sample photos corresponding to the products detected as unqualified can be transmitted to the artificial judgment device for judgment. For example, the unqualified photos can be transmitted to the detection station together with the corresponding qualified photos for display and prompting (e.g., alarm, interactive interface display, etc.). In this way, a second round of judgment can be further performed for less occurring unqualified products, reducing the risk of system misjudgment. Moreover, the photos that are still unqualified after artificial judgment and the photos that are detected as unqualified but judged as qualified by artificial judgment can all be used as new training samples to be fed into the appearance detection model for further iterative training.

[0046] In some embodiments, the labeling process of the method 100 described above can include establishing a classification table, i.e., establishing a classification table for each part to be detected of the product, to classify sample photos of the same type of product for training of the appearance detection model. For example, taking the x and y models of vehicles in the above example of the vehicle production line as an example, the parts to be detected of the x model of vehicles may, for example, include wheels, lights, and windows, and the parts to be detected of the y model of vehicles may, for example, also include wheels, lights, and windows, then 6 classifications (e.g., 6 folders are established) can be included: x model wheels, x model lights, x model windows, y model wheels, y model lights, and y model windows, each of which includes one or more photos of the part. In this way, the corresponding classification table can include, for example, 6 rows of records corresponding to the 6 classifications, 3 rows for each type of vehicle product, and in each row, the basic situation of the corresponding training sample of the row (e.g., the number of training samples, the parameters of the training model, and other machine learning / deep neural network method related parameters and adaptation amounts) can be recorded and stored. Thus, it can be very convenient to manually organize and globally visualize the production line product, and it is beneficial to modify the detection method of the production line (e.g., merge some classifications, delete some classifications of products that are no longer produced, etc.). In particular, in the same large type but different specific type production line (e.g., the x and y models of vehicles described above both belong to the "vehicle" large type, but there are various models of vehicle production), there may be some different types of products using the same part to be detected. The merging step in this paragraph is especially suitable for such production lines, and because of the increased number of training samples, the further simplification of the process, the model training efficiency, the detection efficiency, and ultimately the production efficiency of the production line are further improved.

[0047] On this basis, the training of the appearance detection model further comprises: in response to different types of products in one or more types of products on the production line having different sample photos for different to-be-detected parts, establishing a classification for each to-be-detected part of each product type; and in response to different types of products in one or more types of products on the production line having the same sample photos for different to-be-detected parts, merging the classifications of the same sample photos. That is, if the types of products are different and the types of to-be-detected parts are different, the classification and / or the establishment of the classification table are still performed according to the process of any of the above embodiments. However, in the case where the product types are different but a certain to-be-detected part is the same, the to-be-detected part actually does not need to be divided into multiple categories, and merging the to-be-detected part into one category is also beneficial to increasing the number of training samples, improving the training effect, and further improving the detection accuracy. For example, taking the x model and y model vehicles of the above automobile production line example as an example, the to-be-detected parts of the x model vehicle may, for example, include wheels, lights, and windows, and the to-be-detected parts of the y model vehicle may, for example, also include wheels, lights, and windows. However, in fact, the lights used by the x model and y model vehicles may be the same light. Based on this, when it is determined that the to-be-detected parts are relatively close after the comparison step in the training process, the merging of the categories is automatically performed, that is, the 2 classifications of the lights of the x model and y model vehicles are merged into 1, so as to increase the number of training samples. The comparison step described above can be a sub-step of the embedding training algorithm, that is, the comparison of the samples between the models is automatically performed in the training process, and a reasonable similarity threshold (for example, it can be set that if the similarity is greater than 70%, it can be considered that “different types of products have the same sample photos for different to-be-detected parts”) can be set to determine whether to merge.

[0048] Of course, similarly, after the comparison step described above, the determination of the system can also be transmitted to the manual determination for determination. That is, the method further comprises: in response to different types of products in one or more types of products on the production line having different sample photos for different to-be-detected parts, establishing a classification for each to-be-detected part of each product type; and in response to different types of products in one or more types of products on the production line having the same sample photos for different to-be-detected parts, transmitting the same sample photos to a manual determination device for determination, and merging the classifications of the same sample photos in response to the determination passing. In this way, a second round of determination can be further performed for “different types of products but having the same to-be-detected parts” which may occur less frequently, and if the manual determination still allows merging, the merging is performed, which can reduce the risk of system error merging the classifications.

[0049] As can be seen, most of the above embodiments can be carried out automatically, but in some embodiments, artificial judgment is also used. Regardless, even if artificial judgment is involved, the present process is highly automated, and the work performed by the artificial judgment device is more specialized and single (e.g., only needs to judge whether it is unqualified or whether it is classified and merged), and the entire production detection process is more efficient.

[0050] Figure 2 A module schematic diagram of the production line product appearance detection system 200 according to some embodiments is shown. Each module can correspond to each embodiment described above with respect to the method 100 to implement the details described above with respect to each step of the method 100, and thus in the following, some content will not be described again. In some embodiments, the production line product appearance detection system 200 can include an acquisition device 210, a memory 220, and a processor 230. Figure 2 In some embodiments, the production line product appearance detection system 200 can include an acquisition device 210, a memory 220, and a processor 230.

[0051] The acquisition device 210 acquires one or more sample photos of one or more to-be-detected parts of the product on the detection station and label information of the product. For example, the acquisition device can include a photo acquisition unit (e.g., a camera) for acquiring sample photos and a scanning unit for scanning to obtain the label information of the product. Of course, the scanning unit can also be a separate device from the acquisition device. In some examples, the acquisition device 210 only includes a photo acquisition unit, and the photo acquisition unit takes a photo of the label to allow the system to identify the label in the photo.

[0052] The memory 220 can store instructions that, when executed by the processor 230, perform the following operations: labeling one or more sample photos based on label information of the product; training an appearance detection model for the type of the product using the labeled one or more sample photos as training samples; and detecting the appearance of subsequent products of the same type of the production line using the appearance detection model.

[0053] In some embodiments, the processor 230 is further configured to execute the instructions stored in the memory 220 to iteratively train the appearance detection model using the sample photos of the detected products.

[0054] In some embodiments, the production line product appearance detection system 200 can further include an artificial judgment device (not shown). Alternatively, the system 200 does not include an artificial judgment device, and the artificial judgment device is provided separately from the system 200. The processor 230 is further configured to execute the instructions stored in the memory 220 to, in response to the sample photo of the subsequent product being detected as unqualified, transmit the sample photo corresponding to the detected unqualified product to the artificial judgment device for judgment.

[0055] In some embodiments, the processor 230 is further configured to execute the instructions stored in the memory 220 to establish a classification table for each part to be detected of the product, to classify sample photos of the same type of product, for training of the appearance detection model.

[0056] In some embodiments, the processor 230 is further configured to execute the instructions stored in the memory 220 to, in response to different parts to be detected of different types of products in the one or more types of products on the production line having different sample photos, establish a classification for the sample photos of each part to be detected of each type of product; and in response to different parts to be detected of different types of products in the one or more types of products on the production line having the same sample photos, merge the classifications of the same sample photos.

[0057] In some embodiments, the production line product appearance detection system 200 can further include a manual judgment device (not shown). Alternatively, the system 200 does not include a manual judgment device, which is provided separately from the system 200. The processor 230 is further configured to execute the instructions stored in the memory 220 to, in response to different parts to be detected of different types of products in the one or more types of products on the production line having different sample photos, establish a classification for the sample photos of each part to be detected of each type of product; and in response to different parts to be detected of different types of products in the one or more types of products on the production line having the same sample photos, transmit the same sample photos to the manual judgment device for judgment, and in response to the judgment passing, merge the classifications of the same sample photos.

[0058] The manual judgment device described above may, for example, include a human-computer interaction interface on or near the detection station, which displays the content (e.g. sample photos) to be detected by the worker for judgment. Of course, the manual judgment device can also be a mobile device (e.g. a mobile phone, a tablet, a notebook computer, etc.) installed with a manual judgment program, for remote judgment by the worker.

[0059] According to yet another aspect of the present application, a computer readable storage medium is provided. The computer readable storage medium stores instructions which, when executed by a processor / processing device, implement the method 100 according to any of the preceding embodiments.

[0060] The memory or the computer readable medium mentioned above includes various types of computer storage media that can be accessed by a general purpose or special purpose computer. For example, computer readable media can include computer storage media such as RAM, ROM, EEPROM, E2PROM, solid state drives (SSDs), floppy disks, CD-ROMs, hard disks, other magnetic, optical, physical, or organic storage devices, or any other medium from which a general purpose or special purpose computer can read instructions or data. The foregoing exemplary storage media are intended to be illustrative of the wide scope of storage media available to and readable by a general purpose or special purpose computer and that the exemplary storage media expressly listed herein should not be construed as a limitation upon the scope of computer readable media or computer readable storage media available to and readable by a general purpose or special purpose computer. Combinations of the above should also be included within the scope of computer readable media.

[0061] The memory and the processor can be connected by wire or wirelessly to enable transmission or interaction of data. For example, these components can be electrically connected to each other through one or more communication buses, or connected through wireless communication. The memory can be, for example, the computer readable storage medium described above. The processor can have a signal processing capability processing module, such as a general purpose processor, including a central processing unit CPU, a network processor NP, etc.; it can also be a digital signal processor DSP, an application specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0062] The above merely provides a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Those skilled in the art can think of other feasible changes or replacements according to the technical range disclosed in the present application, and such changes or replacements are all included in the protection scope of the present application. The embodiments and features in the embodiments can be combined with each other without conflict. The protection scope of the present application is subject to the description of the claims.

Claims

1. A method of detecting the appearance of a product on a production line, characterized in that, The production line includes an inspection station, the production line has one or more types of products, each type has one or more products, and each type of product also has one or more to-be-inspected parts, and the method comprises the following steps: Obtaining one or more sample photos of one or more to-be-inspected parts of the products on the inspection station and label information of the products; Based on the label information of the products, the one or more sample photos are labeled, and the labeling further comprises establishing a classification table for each to-be-inspected part of the products to classify the sample photos of the same type of products for training of an appearance detection model; Using the labeled one or more sample photos as training samples to train an appearance detection model for the type of products; and Using the appearance detection model to detect the appearance of subsequent products of the same type on the production line, Wherein, the training of the appearance detection model further comprises: In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-inspected part of each product type; and In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having the same sample photos, merging the classification of the same sample photos. The method further comprises:

2. The method of claim 1, wherein, Using the sample photos of the inspected products to iteratively train the appearance detection model. The method further comprises:

3. The method according to claim 1 or 2, characterized in that, In response to the sample photos of the subsequent products being detected as unqualified, transmitting the sample photos corresponding to the unqualified products to a manual judgment device for judgment. The method further comprises:

4. The method of claim 1, wherein, In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-inspected part of each product type; and In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having the same sample photos, transmitting the same sample photos to a manual judgment device for judgment, and merging the classification of the same sample photos in response to the judgment passing. The production line includes an inspection station, the production line has one or more types of products, each type has one or more products, and each type of product also has one or more to-be-inspected parts, and the system comprises:

5. A production line product appearance inspection system characterized by comprising: An acquisition device that acquires one or more sample photos of one or more to-be-inspected parts of the products on the inspection station and label information of the products; A memory storing instructions; and A processor configured to execute the instructions to perform the following operations: Based on the label information of the products, the one or more sample photos are labeled; Using the labeled one or more sample photos as training samples to train an appearance detection model for the type of products; and Using the appearance detection model to detect the appearance of subsequent products of the same type on the production line, Wherein, the training of the appearance detection model further comprises: In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each to-be-inspected part of each product type; and In response to different to-be-inspected parts of different types of products in one or more types of products on the production line having the same sample photos, merging the classification of the same sample photos. detecting, by the appearance detection model, appearances of subsequent products of the same type on the production line, wherein the processor is further configured to execute the instructions to perform: establishing a classification table for each part to be detected of the product, to classify the sample photos of products of the same type, for training of the appearance detection model, in response to different parts to be detected of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each part to be detected of each product type; and in response to different parts to be detected of different types of products in one or more types of products on the production line having the same sample photos, merging the classifications of the same sample photos.

6. The system of claim 5, wherein, the processor is further configured to execute the instructions to perform: iteratively training the appearance detection model using the detected sample photos of the products.

7. The system of claim 5 or 6, wherein, the processor is further configured to execute the instructions to perform: in response to the sample photo of the subsequent product being detected as unqualified, transmitting the sample photo corresponding to the detected unqualified product to a manual judgment device for judgment.

8. The system of claim 5, wherein, the processor is further configured to execute the instructions to perform: in response to different parts to be detected of different types of products in one or more types of products on the production line having different sample photos, establishing a classification for the sample photos of each part to be detected of each product type; and in response to different parts to be detected of different types of products in one or more types of products on the production line having the same sample photos, transmitting the same sample photos to a manual judgment device for judgment, and in response to a judgment pass, merging the classifications of the same sample photos.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions which, when executed by a processor, perform the production line product appearance detection method according to any one of claims 1-4.

Images