A weighing method and system based on visual image

Abstract

The present invention relates to the technical field of article weighing detection, and specifically to a weighing method and system based on visual images. By collecting multiple target images, extracting the target area of ​​the target image based on the preprocessed image, and calculating the area of ​​the target area, then obtaining the target weight, inputting the target area, the area of ​​the target area and the target weight into a weighing model to train the weighing model, and obtaining the trained weighing model, finally collecting the target real-time image, inputting the target real-time image into the trained weighing model, obtaining the target predicted unit weight output by the trained weighing model, and judging whether the target predicted unit weight is qualified, the present application extracts the target area in the target image, and sets an area threshold to remove the interference of the invalid area in the target image. Since the present application calculates the area of ​​the target area based on pixel points, the present application can be applied to targets of different shapes, thereby improving the accuracy and reliability of the calculation of the unit weight of the cloth.

Description

Technical Field

[0001] The present invention relates to the technical field of article weighing detection, and in particular to a weighing method and system based on visual images. Background Art

[0002] Fabric is a commonly used material in decoration, including chemical fiber carpets, non-woven wall cloths, linen, nylon, colored tape, flannel and other fabrics. Fabric plays a significant role in decorative display.

[0003] A Chinese patent with announcement number CN114543960B discloses an intelligent scale, system and weighing method based on image recognition, which divides an image of an object into multiple areas and extracts color features and texture features of each area; the extracted color features and texture features are fuzzy processed respectively, and compressed and uploaded to the cloud for feature comparison, and the type of object that matches the weighing object is obtained based on the feature comparison result. However, when measuring the unit weight of the cloth, since the shape of the cloth is not fixed, it is difficult to obtain the actual area of ​​the cloth, and thus it is difficult to obtain an accurate unit weight. Summary of the invention

[0004] The purpose of the present invention is to propose a weighing method and system based on visual images in view of the problems existing in the background technology.

[0005] The technical solution of the present invention:

[0006] On the one hand, the present application provides a weighing method based on visual images, comprising:

[0007] Setting acquisition parameters, acquiring multiple target images, preprocessing the target images to obtain preprocessed images, extracting target regions of the target images based on the preprocessed images, and calculating the areas of the target regions;

[0008] Creating a weighing model, obtaining a target weight, inputting a target region, an area of ​​the target region, and a target weight into the weighing model to train the weighing model, and obtaining a trained weighing model;

[0009] Collect the real-time image of the target, input the real-time image of the target into the trained weighing model, obtain the target predicted unit weight output by the trained weighing model, and determine whether the target predicted unit weight is qualified.

[0010] Preferably, setting acquisition parameters, acquiring multiple target images, preprocessing the target images to obtain preprocessed images, extracting a target area of ​​the target image based on the preprocessed images, and calculating the area of ​​the target area, including:

[0011] Create an image database;

[0012] Set acquisition parameters, acquire multiple target images based on the acquisition parameters, and put all acquired target images into an image database;

[0013] Randomly select a target image from the image database;

[0014] Binarization is performed on the target to obtain a binary image;

[0015] Perform edge detection based on the binary image, extract the target area in the binary image based on the edge detection result, and calculate the area of ​​the target area;

[0016] Return to randomly select a target image from the image database until all images in the image database are selected to obtain the target area of ​​each target image.

[0017] Preferably, edge detection is performed based on the binary image, a target region in the binary image is extracted in combination with the edge detection result, and the area of ​​the target region is calculated, including:

[0018] Calculate the horizontal gradient and vertical gradient of each pixel in the binary image respectively;

[0019] Combine horizontal and vertical gradients for non-maximum suppression;

[0020] Set the high threshold and low threshold respectively;

[0021] For all pixels in the binary image, determine whether the pixel value of each pixel is greater than a high threshold or less than a low threshold in turn; record the pixel value greater than the high threshold as a strong edge pixel, and record the pixel value less than the low threshold as a weak edge pixel;

[0022] Connect each strong edge pixel point in sequence to obtain the target peripheral curve; the area surrounded by the target peripheral curve is recorded as the target area.

[0023] Preferably, edge detection is performed based on the binary image, and the target area in the binary image is extracted in combination with the edge detection, and the area of ​​the target area is calculated, and the method further includes:

[0024] Obtain the number of pixels in each target area, and calculate the area of ​​each target area based on the number of pixels in the target area using Formula 1;

[0025] Formula 1;

[0026] in, is the area of ​​the target region, is the area of ​​a single pixel, and P is the number of pixels in the target area;

[0027] Set area threshold;

[0028] Determine in turn whether the area of ​​each target area is less than the area threshold;

[0029] If the area of ​​the target region is smaller than the area threshold, the target region is discarded.

[0030] Preferably, creating a weighing model, obtaining a target weight, inputting the target region, the area of ​​the target region, and the target weight into the weighing model to train the weighing model, and obtaining a trained weighing model, comprises:

[0031] Create a weighing model;

[0032] Divide all target images in the image database into training sets and test sets according to random proportions;

[0033] Inputting the training set into the weighing model to train the weighing model, thereby obtaining a trained weighing model;

[0034] Input the test set into the trained weighing model to verify whether the trained weighing model is trained.

[0035] Preferably, the training set is input into the weighing model to train the weighing model, and the trained weighing model is obtained, comprising:

[0036] Randomly select a target image from the training set and obtain the target area and target weight corresponding to the target image;

[0037] Calculate the unit weight corresponding to the target image using formula 2;

[0038] Formula 2;

[0039] Where W is the unit weight corresponding to the target image, H is the target weight of the target image, and S is the target area of ​​the target image;

[0040] Establishing a coupling relationship of target image-target area-unit weight, and inputting the target image, target area, unit weight and the coupling relationship of target image-target area-unit weight into the weighing model as a training sample;

[0041] Return to randomly select a target image from the training set until all target images in the training set are selected to obtain a trained weighing model; the trained weighing model has the ability to automatically calculate the unit weight of the target according to the input target image and target weight.

[0042] Preferably, collecting a real-time image of the target, inputting the real-time image of the target into a trained weighing model, obtaining a target predicted unit weight output by the trained weighing model, and judging whether the target predicted unit weight is qualified include:

[0043] Collect real-time images of the target;

[0044] Input the real-time image of the target into the trained weighing model to obtain the predicted unit weight of the target;

[0045] Determine whether the prediction result is qualified based on the target predicted unit weight.

[0046] Preferably, judging whether the prediction result is qualified based on the target predicted unit weight includes:

[0047] Set the qualified threshold interval;

[0048] Determine whether the target predicted unit weight is within the qualified threshold range;

[0049] If the target predicted unit weight is within the qualified threshold interval, the target predicted unit weight is qualified;

[0050] If the target predicted unit weight is not within the qualified threshold range, the target predicted unit weight is unqualified.

[0051] On the other hand, the present application also provides a visual image-based weighing system, including an acquisition component, a weighing component and a processing component, wherein the acquisition component is used to acquire target images, the weighing component is used to acquire target weight data, and the processing component is used to execute the visual image-based weighing method described in any one of claims 1 to 8, wherein the acquisition component and the weighing component are both communicatively connected to the processing component, the initial data acquired by the acquisition component and the weighing component are input into the processing component, and the unit weight of the target is calculated by the processing component in combination with the initial data.

[0052] Preferably, the acquisition component includes a camera and a light source.

[0053] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0054] By acquiring multiple target images, preprocessing the target images, extracting the target area of ​​the target image based on the preprocessed images, and calculating the area of ​​the target area, then creating a weighing model, obtaining the target weight, inputting the target area, the area of ​​the target area and the target weight into the weighing model to train the weighing model, and obtaining the trained weighing model, finally acquiring the target real-time image, inputting the target real-time image into the trained weighing model, obtaining the target predicted unit weight output by the trained weighing model, and judging whether the target predicted unit weight is qualified, the present application extracts the target area in the target image and sets an area threshold to remove interference from invalid areas in the target image. Since the present application calculates the area of ​​the target area based on pixel points, the present application can be applied to targets of different shapes, thereby improving the accuracy and reliability of the calculation of the unit weight of the cloth. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of a weighing method based on visual images proposed by the present invention;

[0056] Figure 2 A structural schematic diagram of a visual image-based weighing system proposed by the present invention;

[0057] Figure 3 A schematic diagram of a target image collected by a visual image-based weighing method proposed in the present invention;

[0058] Reference numerals: 100, collection component; 200, weighing component; 300, processing component. DETAILED DESCRIPTION

[0059] Embodiment 1, as Figure 1 As shown, a weighing method based on visual images proposed by the present invention includes:

[0060] S100, setting acquisition parameters, acquiring multiple target images, preprocessing the target images to obtain preprocessed images, extracting target regions of the target images based on the preprocessed images, and calculating the areas of the target regions;

[0061] S200, creating a weighing model, obtaining a target weight, inputting a target region, an area of ​​the target region, and a target weight into the weighing model to train the weighing model, and obtaining a trained weighing model;

[0062] S300, collecting a real-time image of the target, inputting the real-time image of the target into a trained weighing model, obtaining a target predicted unit weight output by the trained weighing model, and determining whether the target predicted unit weight is qualified.

[0063] In the present invention, a plurality of target images are collected, the target images are preprocessed, a target region of the target image is extracted based on the preprocessed image, and the area of ​​the target region is calculated, then a weighing model is created, the target weight is obtained, the target region, the area of ​​the target region and the target weight are input into the weighing model to train the weighing model, and a trained weighing model is obtained, and finally a target real-time image is collected, the target real-time image is input into the trained weighing model, the target predicted unit weight output by the trained weighing model is obtained, and it is judged whether the target predicted unit weight is qualified. The present application extracts the target region in the target image and sets an area threshold to remove interference from invalid regions in the target image. Since the present application calculates the area of ​​the target region based on pixel points, the present application can be applied to targets of different shapes, thereby improving the accuracy and reliability of the calculation of the unit weight of the cloth.

[0064] In an optional embodiment, the S100 includes:

[0065] S110, creating an image database;

[0066] S120, setting acquisition parameters, acquiring multiple target images based on the acquisition parameters, and putting all acquired target images into an image database;

[0067] Specifically, the acquisition parameters include an acquisition frequency and an acquisition period, so that multiple target images can be acquired for the same target within the acquisition period;

[0068] S130, randomly selecting a target image from the image database;

[0069] S140, performing binarization processing on the target to obtain a binarized image;

[0070] Specifically, when performing binarization processing, a pixel value threshold needs to be set to record the pixel values ​​of pixels greater than the pixel value threshold as 1, and the pixel values ​​of pixels less than the pixel value threshold as 0, thereby dividing the target image into a white area and a black area;

[0071] S150, performing edge detection based on the binary image, extracting a target region in the binary image based on the edge detection result, and calculating the area of ​​the target region;

[0072] S160, returning to step S130, until all images in the image database are selected, and the target area of ​​each target image is obtained.

[0073] It should be noted that since the acquisition target of the present application is cloth, and the color of the cloth is different from the color of the weighing component, after the target image is acquired, the target image is binarized to emphasize the cloth area in the target image, so as to facilitate the separation of the cloth from the background area of ​​the weighing component in the subsequent steps, thereby extracting the target area and calculating the area of ​​the target area. The target area in the present application is the area occupied by the cloth in the target image.

[0074] In an optional embodiment, the S150 includes:

[0075] S151, respectively calculating the horizontal gradient and vertical gradient of each pixel in the binary image;

[0076] Optionally, before executing step S151, Gaussian filtering may be performed on the binary image to smooth the binary image and reduce noise in the binary image.

[0077] When calculating the horizontal gradient and vertical gradient, the Sboel operator can be used for calculation. The specific steps are as follows:

[0078] K100, calculate the gradient size of each pixel using formula 3;

[0079] Formula 3;

[0080] in, is the gradient size of the pixel, is the image with horizontal edge detection, is the image with longitudinal edge detection;

[0081] K110, calculate the gradient direction by formula 4;

[0082] Formula 4;

[0083] in, is the gradient direction of the pixel, is the image with horizontal edge detection, is the image with longitudinal edge detection;

[0084] S152, combines horizontal gradient and vertical gradient for non-maximum suppression;

[0085] S153, setting a high threshold and a low threshold respectively;

[0086] S154, for all the pixels in the binary image, determine in turn whether the pixel value of each pixel is greater than a high threshold or less than a low threshold; record the pixel value greater than the high threshold as a strong edge pixel, and record the pixel value less than the low threshold as a weak edge pixel;

[0087] S155, sequentially connect each strong edge pixel point to obtain a target peripheral curve; and record the area surrounded by the target peripheral curve as the target area.

[0088] It should be noted that the present application detects the edge of the target area in the target image through the Canny edge detection algorithm. First, the gradient direction of each pixel in the target image is calculated through the Sobel operator, and then non-maximum suppression is performed. Finally, double threshold processing and edge connection are performed to extract the target area from the target image. The target area is the part occupied by the cloth in the target image, and then the area size of the target area is calculated based on the target area.

[0089] In an optional embodiment, the S150 further includes:

[0090] S156, obtaining the number of pixels in each target area, and calculating the area of ​​each target area based on the number of pixels in the target area by using Formula 1;

[0091] Formula 1;

[0092] in, is the area of ​​the target region, is the area of ​​a single pixel, and P is the number of pixels in the target area;

[0093] Specifically, since the distance between the camera of the acquisition component and the top surface of the weighing component remains unchanged, under the premise that the camera resolution remains unchanged, the size of the picture collected by the camera each time is also the same, so the unit pixel area of ​​the collected target image is also unchanged;

[0094] For example, for a 2-megapixel camera, if the overall area of ​​the target image it captures is 1 square meter, then the unit area of ​​each pixel is equal to 2,000,000 / 1,000,000, that is, 0.5 square millimeters. If the total number of pixels in the target area is 1.2 million, then the area of ​​the target area can be calculated by formula 1 to be 0.6 square meters.

[0095] S157, setting an area threshold;

[0096] S158, determining in turn whether the area of ​​each target area is less than an area threshold;

[0097] S159: If the area of ​​the target region is smaller than the area threshold, the target region is discarded.

[0098] It should be noted that since the collection distance between the collection component and the weighing component in the present application is fixed, the area of ​​each pixel point remains unchanged under the premise of keeping the resolution of the collection component unchanged. In this case, it is only necessary to count the number of pixels contained in the target area and multiply the number of pixels by the area of ​​a single pixel point to obtain the area of ​​the target area.

[0099] This application is aimed at the detection of the weight of fabrics, and there may be residual threads or other debris on the surface of the weighing component. In this case, multiple target areas may appear, such as Figure 3 As shown, the black area is the background area of ​​the weighing component, the white area with the largest area is the target area, and the two white areas with smaller areas are thread ends or debris areas. Since the weight of thread ends or debris is extremely small for the whole fabric, if the thread ends or debris are also recorded as the target area, the unit weight including the thread ends or debris is lower than the actual unit weight of the fabric. Therefore, in order to avoid the influence of thread ends or debris on the weight of the fabric, an area threshold is set to eliminate the target area with an area smaller than the area threshold, so that the calculation of the unit weight of the fabric in this application is more accurate.

[0100] Since the actual area of ​​the cloth is calculated by calculating the pixel points of the target area corresponding to the cloth in the present application, the area of ​​the cloth can be accurately obtained for cloths of different shapes, thereby improving the applicability of the present application.

[0101] In an optional embodiment, the S200 includes:

[0102] S210, creating a weighing model;

[0103] S220, dividing all target images in the image database into a training set and a test set according to a random ratio;

[0104] Specifically, when dividing the training and test sets, the proportion of the training set should be higher than that of the test set;

[0105] S230, inputting the training set into the weighing model to train the weighing model, and obtaining a trained weighing model;

[0106] S240, inputting the test set into the trained weighing model to verify whether the trained weighing model is trained.

[0107] It should be noted that the present application creates and trains a weighing model so that the weighing model continuously learns the relationship between the target image, target weight and target unit weight until the weighing model can automatically output the target predicted unit weight based on the input target image.

[0108] In an optional embodiment, the S230 includes:

[0109] S231, randomly selecting a target image from the training set, and obtaining a target area and a target weight corresponding to the target image;

[0110] S232, calculating the unit weight corresponding to the target image by formula 2;

[0111] Formula 2;

[0112] Where W is the unit weight corresponding to the target image, H is the target weight of the target image, and S is the target area of ​​the target image;

[0113] S233, establishing a coupling relationship of target image-target area-unit weight, and inputting the target image, target area, unit weight, and the coupling relationship of target image-target area-unit weight into a weighing model as a training sample;

[0114] S234, return to step S231, until all target images in the training set are selected, and a trained weighing model is obtained; the trained weighing model has the ability to automatically calculate the unit weight of the target according to the input target image and target weight.

[0115] It should be noted that the target unit weight is calculated by combining the area of ​​the target region and the target weight through the weighing model, so that the trained weighing model has the ability to automatically calculate the unit weight of the target according to the input target image and target weight.

[0116] In an optional embodiment, the S300 includes:

[0117] S310, collecting real-time images of the target;

[0118] S320, inputting the real-time image of the target into the trained weighing model to obtain the predicted unit weight of the target;

[0119] S330, judging whether the prediction result is qualified based on the target predicted unit weight.

[0120] It should be noted that after obtaining the trained weighing model, it is only necessary to collect the real-time image of the target and input it into the trained weighing model to obtain the target predicted unit weight, and then it is necessary to verify whether the target predicted unit weight is qualified.

[0121] In an optional embodiment, the S330 includes:

[0122] S331, setting a qualified threshold interval;

[0123] S332, determining whether the target predicted unit weight is within a qualified threshold range;

[0124] S333, if the target predicted unit weight is within the qualified threshold range, the target predicted unit weight is qualified;

[0125] S334, if the target predicted unit weight is not within the qualified threshold range, the target predicted unit weight is unqualified;

[0126] Optionally, an alert may be issued to the user upon detection of a failure to meet the target predicted unit weight.

[0127] It should be noted that when verifying whether the target predicted unit weight is qualified, it is necessary to set the qualified threshold interval first. The qualified threshold interval is an interval range. If the target predicted unit weight is not within this range, it means that there may be a problem with the weighing test, and an alarm needs to be issued to the user in time.

[0128] like Figure 2As shown, the present application also provides a visual image-based weighing system, including an acquisition component 100, a weighing component 200 and a processing component 300, wherein the acquisition component 100 acquires a target image, and the weighing component 200 acquires target weight data, and the processing component 300 executes a visual image-based weighing method as described in any one of the embodiments in Example 1, wherein the acquisition component 100 and the weighing component 200 are both communicatively connected with the processing component 300, and the initial data acquired by the acquisition component 100 and the weighing component 200 are input into the processing component 300, and the unit weight of the target is calculated by the processing component 300 in combination with the initial data.

[0129] It should be noted that after the acquisition component 100 acquires the target image, it needs to be input into the processing component 300 for preprocessing to obtain a preprocessed image. Then, the processing component 300 trains the weighing model based on the preprocessed image and the target weight data input by the weighing component 200 until the weighing model has the ability to automatically calculate the unit weight of the target based on the input target image and target weight.

[0130] In an optional embodiment, the acquisition component 100 includes a camera and a light source.

[0131] It should be noted that the light source is arranged on the upper side of the weighing component 200 and is perpendicular to the top surface of the weighing component 200, so that the light emitted by the light source can vertically illuminate the top surface of the weighing component 200, thereby ensuring that the target image captured by the camera is clear and bright enough.

[0132] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A weighing method based on visual images, characterized in that: include: Setting acquisition parameters, acquiring multiple target images, preprocessing the target images to obtain preprocessed images, extracting target regions of the target images based on the preprocessed images, and calculating the areas of the target regions; Creating a weighing model, obtaining a target weight, inputting a target region, an area of ​​the target region, and a target weight into the weighing model to train the weighing model, and obtaining a trained weighing model; Collecting a real-time image of the target, inputting the real-time image of the target into a trained weighing model, obtaining the target predicted unit weight output by the trained weighing model, and determining whether the target predicted unit weight is qualified; Create an image database; Set acquisition parameters, acquire multiple target images based on the acquisition parameters, and put all acquired target images into an image database; Randomly select a target image from the image database; Binarization is performed on the target to obtain a binary image; Perform edge detection based on the binary image, extract the target area in the binary image based on the edge detection result, and calculate the area of ​​the target area; Return to randomly select a target image from the image database until all images in the image database are selected, and obtain the target area of ​​each target image; The performing edge detection based on the binary image, extracting the target area in the binary image in combination with the edge detection result, and calculating the area of ​​the target area includes: Obtain the number of pixels in each target area, and calculate the area of ​​each target area based on the number of pixels in the target area using Formula 1; S t =P×S p Formula 1; Among them, S t is the area of ​​the target region, S p is the area of ​​a single pixel, and P is the number of pixels in the target area.

2. A weighing method based on visual images according to claim 1, characterized in that: Perform edge detection based on the binary image, extract the target area in the binary image based on the edge detection result, and calculate the area of ​​the target area, including: Calculate the horizontal gradient and vertical gradient of each pixel in the binary image respectively; Combine horizontal and vertical gradients for non-maximum suppression; Set the high threshold and low threshold respectively; For all pixels in the binary image, determine whether the pixel value of each pixel is greater than a high threshold or less than a low threshold in turn; record the pixel value greater than the high threshold as a strong edge pixel, and record the pixel value less than the low threshold as a weak edge pixel; Connect each strong edge pixel point in sequence to obtain the target peripheral curve; the area surrounded by the target peripheral curve is recorded as the target area.

3. A visual image-based weighing method according to claim 2, characterized in that: Perform edge detection based on the binary image, extract the target area in the binary image in combination with edge detection, and calculate the area of ​​the target area, and also include: Set area threshold; Determine in turn whether the area of ​​each target area is less than the area threshold; If the area of ​​the target region is smaller than the area threshold, the target region is discarded.

4. A visual image-based weighing method according to claim 3, characterized in that: A weighing model is created, a target weight is obtained, and a target region, an area of ​​the target region, and a target weight are input into the weighing model to train the weighing model, and a trained weighing model is obtained, including: Create a weighing model; Divide all target images in the image database into training sets and test sets according to random proportions; Inputting the training set into the weighing model to train the weighing model, thereby obtaining a trained weighing model; Input the test set into the trained weighing model to verify whether the trained weighing model is trained.

5. A visual image-based weighing method according to claim 4, characterized in that: The training set is input into the weighing model to train the weighing model, and the trained weighing model is obtained, including: Randomly select a target image from the training set and obtain the target area and target weight corresponding to the target image; Calculate the unit weight corresponding to the target image using formula 2; Where W is the unit weight corresponding to the target image, H is the target weight of the target image, and S is the target area of ​​the target image; Establishing a coupling relationship of target image-target area-unit weight, and inputting the target image, target area, unit weight and the coupling relationship of target image-target area-unit weight into the weighing model as a training sample; Return to randomly select a target image from the training set until all target images in the training set are selected to obtain a trained weighing model; the trained weighing model has the ability to automatically calculate the unit weight of the target according to the input target image and target weight.

6. A visual image-based weighing method according to claim 5, characterized in that: Collecting a real-time image of the target, inputting the real-time image of the target into the trained weighing model, obtaining the target predicted unit weight output by the trained weighing model, and judging whether the target predicted unit weight is qualified, including: Collect real-time images of the target; Input the real-time image of the target into the trained weighing model to obtain the predicted unit weight of the target; Determine whether the prediction result is qualified based on the target predicted unit weight.

7. A visual image-based weighing method according to claim 6, characterized in that: Based on the target predicted unit weight, determine whether the predicted result is qualified, including: Set the qualified threshold interval; Determine whether the target predicted unit weight is within the qualified threshold range; If the target predicted unit weight is within the qualified threshold interval, the target predicted unit weight is qualified; If the target predicted unit weight is not within the qualified threshold range, the target predicted unit weight is unqualified.

8. A visual image-based weighing system, characterized in that: include: An acquisition component, through which a target image is acquired; A weighing component, through which target weight data is collected; A processing component, through which the visual image-based weighing method as described in any one of claims 1 to 7 is executed, the acquisition component and the weighing component are both communicatively connected to the processing component, the initial data collected by the acquisition component and the weighing component are input into the processing component, and the unit weight of the target is calculated by the processing component in combination with the initial data.

9. A visual image-based weighing system according to claim 8, characterized in that: The acquisition component includes a camera and a light source.

Images