A method and device for improving detection accuracy of bottled objects and foreign object detection equipment

Abstract

The application discloses a method for improving detection accuracy of bottled objects, characterized by being applied to an X-ray foreign matter detection device, and comprising the following steps: collecting a fullness curve of an X-ray irradiation detection channel before starting detection of a to-be-detected bottled object; the installation height of an X-ray source needs to satisfy complete imaging of the bottom of the bottled object; selecting a reference point from a fluctuation interval of the fullness curve; determining a normalized energy value according to the energy value of the reference point; correcting the fullness curve according to the normalized energy value; generating a bottled object detection image with a chain plate based on the corrected fullness curve when the to-be-detected bottled object is detected; and identifying the bottled object detection image with the chain plate to determine whether there is foreign matter in the to-be-detected bottled object. According to the application, the bottom of the bottled object can be completely imaged, and the interference of a transmission chain plate is avoided, so that whether there is foreign matter in the bottom of the bottled object can be accurately identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of foreign matter detection, and in particular to a method and device for improving the detection accuracy of bottled products and a foreign matter detection apparatus. BACKGROUND

[0002] When a side-illuminating X-ray machine is used to test bottled beverages, the bottom chain plate is generally shielded, mainly because the chain plate absorbs too much X-ray, resulting in too much interference signal in the generated image, thereby affecting the detection accuracy. Therefore, the conventional approach is to shield the bottom chain plate. However, since the X-ray is a point light source and the chain plate fluctuates during operation, the chain plate and the bottle bottom will coincide during imaging, resulting in the bottle bottom being shielded together, as shown in FIG. a in the drawing, which shows the imaging effect after the bottle bottom is shielded. After the bottle bottom is shielded, the foreign matter in the bottle bottom cannot be detected, and in the production environment, the foreign matter of liquid beverages is basically deposited at the bottom of the bottle, thus causing the foreign matter in the bottle bottom to be missed. Figure 1

[0003] Currently, the industry has two approaches to this side-illuminating X-ray machine. One is to shield the bottom chain plate, and the other is to adjust the position of the X-ray source to keep the center of the light source level with the bottle bottom, as shown in FIG. b in the drawing, thereby alleviating this missed detection situation to some extent. However, since the conveying chain plate fluctuates, such as the chain plate being warped, the light source cannot always be aligned with the bottle bottom, so these current measures cannot completely solve the problem. Figure 2 SUMMARY

[0004] To solve the above technical problems, the present application provides a method and device for improving the detection accuracy of bottled products and a foreign matter detection apparatus.

[0005] Specifically, the technical solutions of the present application are as follows:

[0006] On the one hand, the present application discloses a method for improving the detection accuracy of bottled products, applied to an X-ray foreign matter detection apparatus, the method comprising: collecting a full-scale curve of an X-ray irradiation detection channel before starting detection of a to-be-detected bottled product; the installation height of the X-ray source needs to satisfy complete imaging of the bottle bottom of the bottled product; selecting a reference point from the fluctuation interval of the full-scale curve; determining a normalized energy value according to the energy value of the reference point; correcting the full-scale curve according to the normalized energy value; generating a bottled product detection image with a chain plate based on the corrected full-scale curve during detection of the to-be-detected bottled product; identifying the bottled product detection image with the chain plate to determine whether there is foreign matter in the to-be-detected bottled product.

[0007] ​​In some embodiments, the reference point is determined according to the fullness curve, specifically comprising: receiving a reference point selection instruction of a user, and selecting a corresponding detection point on the fullness curve of the chain plate image as the reference point according to the selection instruction; or comparing the fullness curve with a pre-stored fullness curve of a normal chain plate, obtaining a mutation point on the fullness curve, and taking the mutation point as the reference point.

[0008] In some embodiments, the fullness curve is corrected according to the normalized energy value, specifically comprising: updating the energy values of the detection points in the fluctuation interval to the normalized energy value, and generating a corrected fullness curve.

[0009] In some embodiments, the bottle detection image with the chain plate is identified to determine whether there is a foreign matter in the bottle to be detected, specifically comprising: inputting the bottle detection image into a trained foreign matter identification detection model, wherein the training samples for training the foreign matter identification model include normal bottle detection image samples with the chain plate and abnormal bottle detection image samples with the chain plate; the normal bottle detection image samples with the chain plate are labeled with a transmission chain plate; the abnormal bottle detection image samples with the chain plate are labeled with a transmission chain plate and a foreign matter respectively; and whether there is a foreign matter in the bottle detection image is identified and output.

[0010] In some embodiments, the bottle detection image with the chain plate is used to determine whether there is a foreign matter in the bottle to be detected, specifically comprising: identifying the chain plate in the bottle detection image according to the image characteristics of the chain plate; the image characteristics of the chain plate image include that the chain plate image is a continuous linear image; and ignoring the chain plate in the bottle detection image, identifying whether there is a foreign matter in the bottle detection image.

[0011] In another aspect, the application also provides a device for improving the detection accuracy of a bottle, comprising: a fullness acquisition module, configured to acquire a fullness curve of a transmission chain plate irradiated by X-rays before a bottle to be detected starts detection; the installation height of the X-ray source needs to satisfy that the bottom of the bottle is fully imaged; a reference point selection module, configured to select a reference point from a fluctuation interval of the fullness curve; a normalization determination module, configured to determine a normalized energy value according to the energy value of the reference point; a correction module, configured to correct the fullness curve according to the normalized energy value; an image generation module, configured to generate a bottle detection image with the chain plate based on the corrected fullness curve when the bottle to be detected is detected; and an identification module, configured to identify the bottle detection image with the chain plate to determine whether there is a foreign matter in the bottle to be detected.

[0012] In some embodiments, the reference point selection module specifically comprises an instruction execution submodule or a comparison selection submodule; wherein the instruction execution submodule is configured to receive a reference point selection instruction of a user, and select a corresponding detection point on the fullness curve of the chain imaging as the reference point according to the selection instruction; and the comparison selection submodule is configured to compare the fullness curve with a pre-stored fullness curve of a normal chain, obtain a mutation point on the fullness curve, and take the mutation point as the reference point.

[0013] In some embodiments, the correction module is specifically configured to update the energy values of the detection points in the fluctuation interval to the normalized energy values, and generate a corrected fullness curve.

[0014] In some embodiments, the device for improving the detection accuracy of the bottled product further comprises a model training module configured to train a foreign matter recognition detection model through training samples in a training set, wherein the training sample set comprises a normal bottled product detection image sample with a chain and an abnormal bottled product detection image sample with a chain; the normal bottled product detection image sample with a chain is labeled with a transmission chain; and the abnormal bottled product detection image sample with a chain is labeled with a transmission chain and a foreign matter respectively.

[0015] The recognition module specifically comprises an input submodule configured to input the bottled product detection image into the trained foreign matter recognition detection model; and a recognition output submodule configured to recognize the chain and the foreign matter in the bottled product detection image through the foreign matter recognition detection model.

[0016] Finally, the present application also provides an X-ray foreign matter detection device comprising the device for improving the detection accuracy of the bottled product.

[0017] Compared with the prior art, the present application has at least one of the following beneficial effects:

[0018] 1. The present application lowers the height of the X-ray source, so that the bottom of the bottle can be completely imaged, thereby avoiding the situation that the presence of a foreign matter in the bottle bottom cannot be detected due to incomplete imaging of the bottle bottom. In addition, although the chain is also imaged, the fullness curve will be corrected in advance during imaging processing, the detection points in the fluctuation interval of the fullness curve caused by the uneven transmission chain will be normalized and assigned a normalized energy value, so that the chain signal in the final imaging is strengthened, facilitating subsequent differentiation and recognition of the foreign matter.

[0019] 2, The selection of reference points in the application can adopt the manual selection mode. Since manual selection is very simple and has strong operability, the reference points can be directly displayed on the screen after being determined by the operator. Then, the subsequent operation can be continued after receiving the instruction. Of course, the reference points can also be automatically selected intelligently. Based on the comparison of the pre-set normal fullness curve, the mutation point of the fullness curve is quickly identified, and the mutation point is taken as the reference point, which is simple and fast.

[0020] 3, The application adopts artificial intelligence to perform image recognition. Specifically, the recognition and detection are strengthened by a deep learning method. The training samples mainly adopt a large number of imaging pictures of bottle containers with chain plates collected on site, including X-ray pictures of bottle containers with chain plates but without foreign matters, X-ray pictures of bottle containers with chain plates and foreign matters, and the samples are classified and labeled to strengthen learning to suppress the strong signal of the chain plate and not to miss the foreign matter signal. Through a large number of sample training, the foreign matter recognition model can recognize the chain plate and will not regard the chain plate as a foreign matter, and can recognize the real foreign matter. BRIEF DESCRIPTION OF DRAWINGS

[0021] The above features, technical characteristics, advantages and implementation modes of the application will be further described in the following preferred embodiments in a clear and understandable manner combined with the drawings.

[0022] Figure 1 is an X-ray imaging schematic diagram of a bottle container to be measured after shielding of the chain plate in the prior art;

[0023] Figure 2 is a schematic diagram of the installation position of the emitter for keeping the bottle bottom imaging in the prior art;

[0024] Figure 3 is another X-ray imaging schematic diagram of the bottle container to be measured in the prior art;

[0025] Figure 4 is a flowchart of one embodiment of the method for improving the detection accuracy of the bottle container in the application;

[0026] Figure 5 is a schematic diagram of the installation position of the emitter for ensuring the complete imaging of the bottle bottom in the application;

[0027] Figure 6 is a schematic diagram of the fullness curve before correction;

[0028] Figure 7 is a schematic diagram of the fullness curve after correction;

[0029] Figure 8 is an X-ray imaging schematic diagram of the bottle container with the chain plate after the fullness curve is corrected in the application;

[0030] Figure 9is a structural block diagram of one embodiment of the device for improving detection precision of bottled objects of the application;

[0031] Figure 10 is a structural block diagram of another embodiment of the device for improving detection precision of bottled objects of the application.

[0032] Explanation of reference numerals:

[0033] a--bottle bottom area; 1--X-ray emitter; 2--receiver; 3--conveyor; 4--bottled object; 10--fullness acquisition module; 20--reference point selection module; 30--normalization determination module; 40--correction module; 50--image generation module; 60--recognition module; 21--instruction execution sub-module; 22--comparison and selection sub-module; 61--input sub-module; 62--recognition output sub-module. DETAILED DESCRIPTION

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, specific embodiments of the present application will be described below with reference to the drawings. Obviously, the drawings in the following description only represent some embodiments of the present application, and for those skilled in the art, other drawings can be obtained from these drawings without creative labor, and other embodiments can also be obtained.

[0035] In order to make the drawing simple, only the parts related to the application are shown in each drawing, which does not represent the actual structure of the product. In addition, in order to make the drawing simple and easy to understand, in some drawings, only one of the components with the same structure or function is shown schematically, or only one of them is marked. In this paper, "one" not only means "only one", but also means "more than one".

[0036] It should be further understood that the term "and / or" used in the specification and claims of the present application means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0037] In this paper, it should be noted that unless otherwise specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0038] Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0039] Traditional bottled goods inspection typically uses shielded conveyor belt imaging. However, as mentioned earlier, shielding the conveyor belt imaging also masks part of the bottle bottom. To avoid missing foreign objects at the bottle bottom, our researchers considered using imaging with a conveyor belt. However, another problem arose: the conveyor belt's imaging signal is very strong, and it produces random imaging, such as... Figure 3 As shown in the figure, in the bottle bottom area shown in a, the chain plate causes great interference to the imaging of the bottle bottom. Since the chain plate has a randomizing effect on the signal to be detected, it greatly affects the accuracy of the traditional detection algorithm, thus reducing the detection effect. So, is there a way to detect the strong signal of the chain plate? After the inventor racked his brains and made continuous debugging and verification, he finally designed the following solution.

[0040] In one embodiment, refer to the appendix to the specification. Figure 4 The method for improving the detection accuracy of bottled materials provided by this invention is applied to X-ray foreign object detection equipment, and the method includes:

[0041] S100, before starting the detection of the bottled product to be tested, acquire the full-scale curve of the X-ray irradiation detection channel; the installation height of the X-ray source must be sufficient to ensure complete imaging of the bottom of the bottled product.

[0042] Specifically, the installation height of the X-ray source must ensure complete imaging of the bottom of the bottle. During installation, the test bottle can be continuously placed into the conveyor belt, and the height of the X-ray source can be adjusted to ensure complete imaging of the bottle bottom. Simultaneously, the conveyor belt will also provide imaging. Figure 5 As shown, compared to the traditional installation height of the X-ray source during bottle inspection, this embodiment lowers the X-ray machine to align with the center point, allowing for complete imaging of the bottle bottom and ensuring that the bottom of all test bottles is fully imaged. Then, the test bottle is removed, and the transport chain is run unloaded, acquiring the full-scale curve of the X-ray machine's conveyor belt image.

[0043] S200, Select a reference point from the fluctuation range of the full-scale curve;

[0044] Specifically, when inspecting bottled goods, to ensure smooth transport, the inspection channel typically uses a conveyor belt system. However, unevenness or undulations, such as warped edges, can exist at the joints between the conveyor belts, resulting in fluctuations in the full-scale curve due to these unevennesses. Figure 6 As shown. After determining the fluctuation range from the full-scale curve, a point is selected from the fluctuation range as a reference point. Generally, the starting point of the fluctuation is selected as the reference point. Figure 6The fluctuation range in the diagram is from x1 to x2, with the fluctuation start point x1 taken as the reference point. x2 in the diagram represents the fluctuation end point; however, the value of x2 can also be obtained based on the initial configuration of the detector.

[0045] S300, determine the normalized energy value based on the energy value of the reference point;

[0046] Specifically, the energy value of the reference point can be obtained based on the ordinate value of the reference point in the full-scale curve, and this energy value can be used as the normalized energy value. Of course, a value lower than the energy value of the reference point can also be selected as the normalized energy value, but selecting the energy value of the reference point as the normalized energy value makes the final image of the chain plate easier to identify and distinguish.

[0047] S400, the full-scale curve is corrected based on the normalized energy value;

[0048] Specifically, after determining the normalized energy value, the energy values ​​of all probe points in the interval from x1 to x2 are set to this normalized energy value, thereby obtaining the corrected full-scale curve, such as... Figure 7 As shown. Specifically, step S400 includes:

[0049] Based on the full-scale curve, determine the fluctuation range of the full-scale curve;

[0050] The energy values ​​of the detection points in the fluctuation range are updated to the normalized energy values ​​to generate the corrected full-scale curve.

[0051] S500, based on the corrected full-scale curve, generates a bottle detection image with a chain plate during the detection of the bottled material to be tested;

[0052] Specifically, during the imaging of the bottled substance to be tested, a corrected full-scale curve is used to assist in imaging, generating a curve such as... Figure 8 The image shown is of a bottled product with a chain plate. Figure 8 In the bottle bottom region shown in Figure a, although the chain plate is also imaged at the same time as the bottle bottom, it is clear that after the full-scale curve is corrected, the image of the chain plate is not random and the image is easy to identify (the thick black line at the bottom of region a is the image of the chain plate). The interference to the image of the bottle bottom is greatly reduced. Thus, the chain plate in the image is easier to detect and identify and will not be confused with foreign objects at the bottom of the bottle.

[0053] S600, identify the detection image of the bottled product with chain plate, and determine whether there are foreign objects in the bottled product to be tested.

[0054] Finally, based on the image of the bottled product with the chain plate, an identification and judgment are performed to determine whether there are foreign objects in the bottled product. Although the image of the bottled product (Figure 8 (with chain plates, but compared to the image with chain plates generated without correcting the full-scale curve) Figure 3 Obviously, the chain plate in the bottled material detection image generated by this method is enhanced, making it easier to identify and judge, and less likely to be confused with foreign objects.

[0055] In this embodiment, by enabling the bottom of the bottle to be imaged and removing the imaging noise introduced by the conveyor chain, we lower the X-ray machine to align with the center point to enable the bottom of the bottle to be fully imaged. At the same time, we use the method of normalizing the energy value of the fluctuation detection point to remove imaging noise. Based on this, we can obtain a complete image of the bottle under test so that the image meets the standard that the algorithm can detect.

[0056] In another embodiment of this application, based on the above embodiment, step S200 determines the reference point according to the full-scale curve; specifically, this can be achieved in the following way:

[0057] (1) Manual selection of reference points

[0058] In other words, the collected full-scale curve is displayed on the screen, and the operator can select a reference point from this displayed curve. Generally, the operator will choose the starting point of the fluctuation in the full-scale curve as the reference point. After manually selecting the reference point, the following operations are performed from the perspective of the machine:

[0059] S210, receives the user's reference point selection command;

[0060] S220, according to the selection instruction, select the corresponding detection point as the reference point on the full-scale curve of the chain plate imaging.

[0061] (2) Intelligent selection of reference points

[0062] Specifically, this solution eliminates the need for manual selection of reference points; instead, the machine intelligently selects reference points based on preset rules. For example, the full-scale curve can be compared with the full-scale curve of a pre-stored normal conveyor belt to identify abrupt changes in the full-scale curve, and these abrupt changes can be used as reference points. Here, a normal conveyor belt refers to a perfectly flat conveyor belt without any warping. In reality, after a period of use, conveyor belts will deform to varying degrees, resulting in warping and unevenness, especially at the joints between belts.

[0063] Of course, other selection methods can also be used to select reference points. The reference point does not necessarily have to be the starting point of the fluctuation. Other points in the fluctuation range can also be selected as reference points. However, if the starting point of the fluctuation is selected as the reference point, the final imaging effect will be better and the final image recognition will be easier.

[0064] In another embodiment of this application, based on any of the above embodiments, step S600 identifies the detection image of the bottled material with the chain plate to determine whether there are foreign objects in the bottled material to be tested; specifically, this can be achieved in the following way:

[0065] (1) Image recognition is performed using artificial intelligence.

[0066] Specifically, the first step is to train a foreign object detection model. The training samples for this model include normal bottled goods with chain plates and abnormal bottled goods with chain plates. The normal bottled goods image samples are labeled with the transmission chain plate; the abnormal bottled goods image samples are labeled with both the transmission chain plate and the foreign object. In other words, deep learning is used to enhance detection. The training samples primarily consist of a large number of images of bottled goods with chain plates collected on-site. These images are generated based on the modified full-scale curve described in this application. The sample images are divided into two categories: X-ray images of bottled goods with chain plates but no foreign objects, and X-ray images of bottled goods with chain plates and foreign objects. Reinforcement learning is used through classification and labeling to suppress strong signals from the chain plate while still detecting foreign object signals. Through extensive training with numerous samples, the foreign object detection model can identify the chain plate without mistaking it for a foreign object, thus recognizing the actual foreign object.

[0067] After the model is trained, in actual detection, the generated bottled object detection image is input into the trained foreign object recognition and detection model, and finally the model identifies and outputs whether there is a foreign object in the bottled object detection image.

[0068] (2) Image recognition based on chain plate image features

[0069] Specifically, this solution does not employ a neural network model. Instead, it identifies the chain plate in the bottled product detection image based on the image characteristics of the chain plate. These image characteristics include the chain plate image being a continuous linear image. Although the bottled product detection image generated using the corrected full-scale curve contains the chain plate, it is enhanced. Because the chain plate has a very obvious linear continuity, it is relatively easy to identify. After identifying the chain plate, it is ignored in the bottled product detection image, and the system then identifies whether foreign objects are present in the image. In other words, the chain plate can be quickly identified based on its inherent characteristics, without being confused with foreign objects.

[0070] Based on the same technical concept, another embodiment of this application provides a device for improving the accuracy of bottled product detection, such as... Figure 9 As shown, it includes:

[0071] The full-scale acquisition module 10 is used to acquire the full-scale curve of the X-ray irradiation transmission chain plate before the detection of the bottled product is started; the installation height of the X-ray source must be sufficient to ensure complete imaging of the bottom of the bottled product.

[0072] Specifically, before starting the detection, full-scale data acquisition is required for subsequent image correction. Full-scale data acquisition refers to obtaining the full-scale curve of X-ray irradiation of the transmission chain plate with the X-ray source turned on but without any objects placed on the conveyor belt. In this embodiment, the installation height of the X-ray source must ensure complete imaging of the bottom of the bottle; that is, the center point of the X-ray source cannot be higher than the bottom of the bottle. However, it can be slightly lower, for example, the center of the X-ray source can be 1 cm below the surface of the transmission chain plate. This ensures that the bottom of the bottle is completely imaged during subsequent imaging. Only by ensuring complete imaging of the bottle bottom can the bottom of the bottle be detected without missing any areas.

[0073] The reference point selection module 20 is used to select a reference point from the fluctuation range of the full-scale curve;

[0074] The normalization determination module 30 is used to determine the normalized energy value based on the energy value of the reference point;

[0075] The correction module 40 is used to correct the full-scale curve based on the normalized energy value.

[0076] The image generation module 50 is used to generate a bottle detection image with a chain plate during the detection of the bottled product, based on the corrected full-scale curve.

[0077] The identification module 60 is used to identify the detection image of the bottled product with the chain plate and determine whether there are foreign objects in the bottled product to be tested.

[0078] In this embodiment, the installation height of the X-ray source ensures complete imaging of the bottle bottom. However, this also introduces interference from the chain plate. While the bottle bottom is being imaged, a portion of the chain plate is also being imaged, especially if the chain plate is uneven, has undulations, or warped edges, resulting in greater interference. However, the device used in this embodiment to improve the accuracy of bottled material detection can fundamentally correct the full-scale curve, reducing the impact of chain plate undulations on the full-scale curve's fluctuations. The energy value of the reference point is used to normalize the energy value of the detection point within the fluctuation range. Therefore, although there is chain plate imaging interference, the imaging of the bottled material based on the corrected full-scale curve is easily identifiable and less likely to be confused with foreign objects. This allows for better differentiation and identification of interference within the bottled material, particularly accurately identifying the presence of foreign objects at the bottom of the bottle.

[0079] Another embodiment of the device for improving the accuracy of bottled product detection in this application, such as Figure 10As shown, based on the above-described device embodiment, the reference point selection module 20 specifically includes:

[0080] The instruction execution submodule 21 is used to receive the user's reference point selection instruction and select the corresponding detection point as the reference point on the full-scale curve of the chain plate imaging according to the selection instruction.

[0081] or

[0082] The comparison and selection submodule 22 is used to compare the full-scale curve with the pre-stored full-scale curve of a normal chain plate, obtain the abrupt change point on the full-scale curve, and use the abrupt change point as a reference point.

[0083] Specifically, reference points can be selected manually or intelligently. Manual selection requires displaying the acquired full-scale curve on a screen, and then the operator or technician selects a reference point on that curve. Generally, the starting point of the curve's fluctuation is chosen as the reference point, that is, the point of abrupt change in the full-scale curve. Figure 6 The x1 point in the diagram. Of course, this point can also be selected automatically and intelligently. Automatic selection requires pre-storing the full-scale curve of a normal chain plate, and then comparing the currently collected full-scale curve with the pre-stored full-scale curve to find the abrupt change point, and thus using the abrupt change point as a reference point.

[0084] In another embodiment, based on any of the above device embodiments, the correction module 40 is specifically used to update the energy values ​​of the detection points in the fluctuation range to the normalized energy values, thereby generating the corrected full-scale curve.

[0085] Specifically, after determining the reference point, the energy value of the reference point is obtained as the normalized energy value, and then the energy values ​​of the detection points in the fluctuation range of the full-scale curve are updated to the normalized energy value.

[0086] In another embodiment of this application, based on any of the above-described device embodiments, the device for improving the accuracy of bottled product detection further includes:

[0087] The model training module 00 is used to train a foreign object recognition and detection model using training samples in a training set. The training sample set includes normal bottled goods detection image samples with chain plates and abnormal bottled goods detection image samples with chain plates. The normal bottled goods detection image samples with chain plates are labeled with a transmission chain plate. The abnormal bottled goods detection image samples with chain plates are labeled with both a transmission chain plate and a foreign object.

[0088] The identification module 60 specifically includes:

[0089] Input submodule 61 is used to input the bottled object detection image into the trained foreign object recognition and detection model;

[0090] The identification output submodule 62 is used to identify the chain plate and foreign objects in the bottled material detection image through the foreign object identification and detection model.

[0091] This embodiment employs artificial intelligence to identify foreign objects in images, utilizing deep learning to enhance detection. Specifically, a foreign object detection model is first trained to identify chain plates and foreign objects in bottled product images. The training samples for this model primarily consist of numerous on-site images of bottled products with chain plates, generated based on the modified full-scale curve described in this application. The sample images are divided into two categories: X-ray images of bottles with chain plates but no foreign objects, and X-ray images of bottles with chain plates and foreign objects. Reinforcement learning is used through classification and labeling to suppress strong chain plate signals while still detecting foreign object signals. Through extensive training with numerous samples, the foreign object detection model can now identify chain plates without mistaking them for foreign objects, thus distinguishing genuine foreign objects. Furthermore, due to the modified full-scale curve, subsequent chain plate imaging is more prominent, making it easier to identify and less prone to confusion with other foreign objects.

[0092] Finally, this application also provides an X-ray foreign object detection device, including the apparatus for improving the detection accuracy of bottled materials as described in any of the above embodiments.

[0093] Specifically, by integrating the device for improving the detection accuracy of bottled materials according to any of the above embodiments into an X-ray foreign object detection device, bottled materials can be detected accurately and quickly.

[0094] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for improving the detection accuracy of a bottled object, characterized by, The method is applied to an X-ray foreign matter detection device, and the method comprises: Before starting detection of the to-be-detected bottled object, a full-scale curve of an X-ray irradiation detection channel is collected, the full-scale curve being a full-scale curve obtained by X-ray irradiation of a transmission chain plate with no object placed on the transmission belt, and a height of a ray source of the X-ray being required to satisfy complete imaging of a bottle bottom of the bottled object; A reference point is selected from a fluctuation interval of the full-scale curve; A normalized energy value is determined according to an energy value of the reference point; The full-scale curve is corrected according to the normalized energy value, specifically including updating energy values of detection points in the fluctuation interval to the normalized energy value to generate a corrected full-scale curve; Based on the corrected full-scale curve, a bottled object detection image with a chain plate is generated during detection of the to-be-detected bottled object; Whether there is foreign matter in the to-be-detected bottled object is judged by recognizing the bottled object detection image with the chain plate.

2. The method of claim 1, wherein, The reference point is selected from the fluctuation interval of the full-scale curve, specifically including: A reference point selection instruction of a user is received, and a corresponding detection point in the fluctuation interval of the full-scale curve of the chain plate imaging is selected as the reference point according to the selection instruction; Or The full-scale curve is compared with a pre-stored full-scale curve of a normal chain plate, a mutation point on the full-scale curve is obtained, and the mutation point is taken as the reference point.

3. The method of claim 1, wherein, Whether there is foreign matter in the to-be-detected bottled object is judged by recognizing the bottled object detection image with the chain plate, specifically including: The bottled object detection image is input into a trained foreign matter recognition detection model, and training samples for training the foreign matter recognition model include normal bottled object detection image samples with a chain plate and abnormal bottled object detection image samples with a chain plate; the normal bottled object detection image samples with the chain plate are labeled with the transmission chain plate; and the abnormal bottled object detection image samples with the chain plate are labeled with the transmission chain plate and the foreign matter respectively; Whether there is foreign matter in the bottled object detection image is recognized and output.

4. The method of claim 1, wherein, Whether there is foreign matter in the to-be-detected bottled object is judged by recognizing the bottled object detection image with the chain plate, specifically including: According to image characteristics of the chain plate, the chain plate in the bottled object detection image is recognized; the image characteristics of the chain plate image include that the chain plate image is a continuous linear image; The chain plate in the bottled object detection image is ignored, and whether there is foreign matter in the bottled object detection image is recognized.

5. A device for improving the detection accuracy of a bottled object, characterized by comprising: The method comprises: A full-scale collection module is configured to collect a full-scale curve of X-ray irradiation of a transmission chain plate before starting detection of a to-be-detected bottled object, the full-scale curve being a full-scale curve obtained by X-ray irradiation of the transmission chain plate with no object placed on the transmission belt, and a height of a ray source of the X-ray being required to satisfy complete imaging of a bottle bottom of the bottled object; A reference point selection module is configured to select a reference point from a fluctuation interval of the full-scale curve; A normalization determination module is configured to determine a normalized energy value according to an energy value of the reference point; A correction module is configured to correct the full-scale curve according to the normalized energy value. The correction module is specifically configured to update the energy values of the detection points in the fluctuation interval to the normalized energy values, and generate a corrected fullness curve. An image generation module is configured to generate a bottle detection image with a chain plate based on the corrected fullness curve when the bottle to be detected is detected. An identification module is configured to identify the bottle detection image with the chain plate, and determine whether there is a foreign object in the bottle to be detected.

6. The device for improving the detection accuracy of a bottle according to claim 5, wherein The reference point selection module specifically includes: An instruction execution sub-module is configured to receive a reference point selection instruction of a user, and select a corresponding detection point on the fullness curve of the chain plate imaging as a reference point according to the selection instruction. Or A comparison and selection sub-module is configured to compare the fullness curve with a pre-stored fullness curve of a normal chain plate, obtain a mutation point on the fullness curve, and select the mutation point as the reference point.

7. The device for improving the detection accuracy of a bottle according to claim 5, wherein Further comprising: A model training module is configured to train a foreign object identification and detection model by using training samples in a training set, wherein the training samples include bottle detection image samples with a normal chain plate and bottle detection image samples with an abnormal chain plate, and the bottle detection image samples with the normal chain plate are labeled with a transmission chain plate. The bottle detection image samples with the abnormal chain plate are labeled with a transmission chain plate and a foreign object respectively. The identification module specifically includes: An input sub-module is configured to input the bottle detection image into the trained foreign object identification and detection model. An identification output sub-module is configured to identify the chain plate and the foreign object in the bottle detection image by using the foreign object identification and detection model.

8. An x-ray foreign object detection apparatus, characterized by, The device for improving the detection precision of the bottle is described in any one of claims 5-7.

Images