Injection molding production line real-time monitoring method and system based on machine vision

Abstract

The present application relates to the field of image processing, in particular to a real-time monitoring method and system for injection molding production line based on machine vision, which divides an image into multiple analysis regions, calculates the region noise degree representing the brightness instability of the region, calculates the pixel noise confidence in combination with the deviation degree of the pixel gradient and its neighborhood, and then fuses the multi-dimensional information to obtain the real noise probability of each pixel. Finally, the smoothing strength of bilateral filtering is adaptively adjusted based on the probability, and strong smoothing is applied to the pixels with high noise probability. The present application can accurately identify and process high light noise, effectively denoise while retaining the real defect details to the maximum, and significantly improve the monitoring image quality and defect detection accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing, in particular to a real-time monitoring method and system for injection molding production line based on machine vision. BACKGROUND

[0002] Injection molding is a commonly used technology for producing plastic products in modern manufacturing industry. In the production process, in order to ensure the quality of the final product, strict quality detection of the surface of the injection molded part is required to timely find and eliminate unqualified products with defects such as scratches, bubbles, and material shortages. With the improvement of industrial automation level, automatic detection systems based on machine vision have become the mainstream technical solution to replace traditional manual visual inspection due to their advantages of high efficiency, objectivity, and non-contact. However, when applying machine vision technology to detect injection molded parts, the quality of image acquisition is challenged by multiple factors. The surface of the injection molded part is often smooth and has certain light reflection characteristics. Under the complex lighting conditions in the industrial field, it is easy to form a "highlight" area with extremely high brightness in the image. This highlight area in the image is characterized by strong brightness noise, and its characteristics are sometimes similar to real defects (such as bright scratches), or it can mask small defects underneath due to its extremely high brightness. This noise caused by uneven lighting is one of the main reasons for the decline in the performance of subsequent defect detection algorithms, resulting in false positives or false negatives. Therefore, how to effectively filter out the highlights and other lighting artifacts while preserving the details of real defects is a key technical problem in this field.

[0003] To solve the above-mentioned image noise problem, a bilateral filtering algorithm is often used in the prior art. This algorithm can smooth the image and suppress noise while preserving edge details. The basic principle is to consider the spatial distance and brightness similarity of pixels, and give higher weights to pixels that are both close and similar to the center point. However, the standard bilateral filtering algorithm has inherent defects when dealing with the highlight problem on the surface of the injection molded part: since the edge of the highlight area also has a large brightness gradient, the traditional bilateral filtering algorithm will misjudge it as a real "edge" that needs to be preserved, and thus does not perform effective smoothing on it. This results in the highlight noise being completely preserved in the image, which not only fails to achieve the purpose of denoising, but also continuously interferes with subsequent defect segmentation and recognition, and cannot meet the needs of high-precision detection. SUMMARY

[0004] In view of the misjudgment problem of the above-mentioned traditional bilateral filtering algorithm, in a first aspect, the present application proposes a real-time monitoring method for injection molding production line based on machine vision, comprising: acquiring a surface image of an injection molded part, and segmenting the surface image into a plurality of analysis regions; calculating the region noise degree of each analysis region, which is used to represent the brightness fluctuation intensity and brightness range of the analysis region; calculating the pixel noise confidence of the pixel points in each analysis region, which is used to measure the deviation degree of the gradient of a single pixel point from the overall gradient feature of the analysis region; based on the region noise degree and the pixel noise confidence, and combining the gradient direction consistency of each analysis region, calculating the real noise probability of each pixel point; based on the real noise probability, acquiring a corrected smoothing intensity factor, which is used to smooth the noise in the image; and adaptively adjusting the smoothing intensity of each pixel point by the bilateral filtering algorithm according to the corrected smoothing intensity factor, to obtain a denoised image.

[0005] The present application segments the image into analysis regions, and calculates the region macro noise (region noise degree) and pixel micro abnormality (pixel noise confidence) in layers, and finally fuses them into a unified noise probability to intelligently guide the strength of bilateral filtering. Compared with the filtering method in the prior art which uses fixed parameters or single-dimensional information for judgment, the present application can dynamically and accurately adaptively process according to complex image content (especially when high light and real defects coexist). This method effectively solves the problem that the traditional filtering algorithm easily misjudges the artifacts as edges in the high light area and retains noise, significantly improves the suppression effect of the surface noise of the injection molded part under complex lighting conditions and the protection ability of the real defect details, and finally improves the accuracy and robustness of defect detection.

[0006] Further, the calculation method of the real noise probability is specifically:

[0007] ;

[0008] wherein represents the real noise probability of the pixel point ; represents the region noise degree of the i-th analysis region in the j-th frame image; represents the pixel noise confidence of the normalized pixel point ; represents the consistency coefficient of the i-th analysis region.

[0009] The present application uses a probability model to combine the region macro noise level , the pixel microstructure abnormality , and the region structure consistency​​​ The evidence of the three dimensions is fused to realize the quantitative evaluation of the possibility that the pixel belongs to noise. Compared with the logic of the prior art that usually adopts independent hard threshold to judge different features, the fusion decision mechanism of the application is more robust and fine. It integrates multiple information, so that the final judgment result is more comprehensive and accurate than the single feature threshold method, greatly improving the confidence of noise recognition.

[0010] Further, the calculation method of the pixel noise confidence is specifically:

[0011] ;

[0012] Wherein represents the pixel noise confidence of the pixel point ; represents the gradient direction of the pixel point ; represents the vector mean of the gradient direction in the th analysis region; represents the gradient amplitude of the pixel point ; represents the gradient amplitude mean of all pixel points in the th analysis region; represents an adjustment coefficient; represents the maximum gradient amplitude in the th frame image.

[0013] The application quantifies the deviation degree of the gradient feature of a single pixel from the overall gradient trend of the region where it is located by constructing the pixel noise confidence. Compared with the prior art that only relies on the gradient amplitude to judge whether the pixel is an edge point, the application adds the consideration of gradient direction consistency. This makes the algorithm able to effectively distinguish the structural edges formed by real defects (such as scratches), which have relatively consistent gradient directions, from isolated noise points formed by highlights or random noise, which have chaotic gradient directions, thereby significantly reducing the risk of misjudging noise as effective details.

[0014] Further, the calculation method of the consistency coefficient is specifically:

[0015] ;

[0016] Wherein represents the consistency coefficient of the th analysis region; represents the gradient direction of the pixel point in the th frame image; represents the total number of pixels in the th analysis region.

[0017] Furthermore, the method for calculating the noise level in the area is as follows:

[0018] ;

[0019] in Indicates the first The first frame of the image The regional noise level of each analysis area; and These are the variance and mean of the brightness values ​​of all pixels within the analysis area, respectively. and These are the variance and mean of the brightness values ​​of the pixels within 8 neighborhoods surrounding the cluster center point of the analysis region, respectively; the cluster center refers to the cluster center in each analysis region when performing superpixel segmentation of the image in step S1. and These represent the maximum and minimum brightness values ​​within the analysis area, respectively. This represents the normalization function.

[0020] This invention integrates three dimensions—overall brightness fluctuation, core area brightness fluctuation, and brightness dynamic range—to comprehensively assess the visual "clutter" of a region by evaluating its regional noise level. Compared to existing technologies that assess noise levels solely through variance or brightness range, this invention's multi-dimensional indicators provide a more comprehensive and reliable measure of noise interference in a region. This method can more accurately identify complex noise areas such as high-light reflections caused by uneven illumination, avoiding the bias and misjudgment that can result from a single indicator, and providing a more reliable macroscopic basis for subsequent pixel-level analysis.

[0021] Furthermore, the cluster centers of the analysis region are obtained through a superpixel segmentation algorithm based on simple linear iterative clustering.

[0022] Furthermore, the specific method for calculating the modified smoothing intensity factor is as follows:

[0023] ;

[0024] in Represents pixels Corrected smoothing intensity factor; This represents the baseline smoothing intensity factor for traditional bilateral filtering; Represents pixels The true noise probability; This represents the intensity amplification factor, used to control the maximum increase in smoothing intensity; This represents the non-linear adjustment index, used to enhance the filtering effect on pixels with high noise probability.

[0025] The application realizes dynamic and nonlinear adjustment of bilateral filtering strength by constructing a modified smoothing strength factor positively correlated with noise probability. Compared with the fixed smoothing parameter of bilateral filtering in the prior art, which cannot balance the noise removal effect and detail preservation in complex scenes, the application realizes differentiated processing of different pixels. Strong smoothing is applied to pixels with high noise probability, which can effectively remove stubborn noise, and slight influence is applied to pixels with low noise probability, thereby effectively removing noise while protecting the real edges and defect details of the image to the maximum extent.

[0026] Further, for noise points with a real noise probability less than a set threshold, a reference smoothing strength factor of the traditional bilateral filtering is used for smoothing processing.

[0027] Further, the surface map of the injection molded part is obtained by: obtaining a real-time video stream from an industrial camera deployed on the injection molding production line; performing preprocessing of de-jittering and illumination correction on the video stream; and converting the preprocessed video frame into a luminance image to obtain the surface map of the injection molded part.

[0028] In a second aspect, the application provides a machine vision-based real-time monitoring system for an injection molding production line, comprising a processor and a memory, wherein the memory stores computer program instructions which, when executed by the processor, implement the machine vision-based real-time monitoring method for an injection molding production line of the application.

[0029] The technical effects of the application are:

[0030] The application discards the traditional mode of relying on a single indicator to judge noise and innovatively constructs a noise evaluation system with two levels of "region" and "pixel". First, the "region noise degree" is used to macroscopically evaluate the brightness confusion degree of a region, and then the "pixel noise confidence" is used to microscopically measure the structural prominence of a single pixel. Then, the above two-layer evaluation results and information such as region gradient consistency are fused, and the accurate probability value of each pixel belonging to noise is calculated through a probability model. Finally, a direct mapping from probability to strength is established: the calculated noise probability is directly linked to the smoothing strength of the filter, and a nonlinear adaptive adjustment function is established. This allows the filtering strength to be fine-tuned for each pixel, realizing pixel-by-pixel adaptive filtering. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 is a flow chart of the machine vision-based real-time monitoring method for an injection molding production line in the embodiment of the application;

[0032] Figure 2 is a structural block diagram of the machine vision-based real-time monitoring system for an injection molding production line in the embodiment of the application. DETAILED DESCRIPTION

[0033] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of the present application.

[0034] The specific embodiments of the present application will be described in detail below with reference to the drawings.

[0035] Embodiment of the injection molding production line real-time monitoring method based on machine vision:

[0036] As shown in Figure 1 The injection molding production line real-time monitoring method based on machine vision of the present application comprises:

[0037] S1, obtaining the surface image of the injection molded part and dividing it into multiple analysis regions.

[0038] The purpose of this step is to obtain the surface image of the injection molded part from the real-time monitoring video of the injection molding production line, and to divide the image into multiple local regions, quantify the noise interference degree of each region, and provide a basis for subsequent adaptive noise reduction processing. In the injection molding production environment, factors such as light changes, equipment vibrations, or dust in the air may cause the image noise to exhibit different characteristics in different regions.

[0039] In specific implementation, first, a real-time video stream is obtained from an industrial camera deployed on the injection molding production line (for example, at the station where the product is ejected and transferred by the conveyor belt). For example, assuming that the video resolution is 1080p (1920x1080 pixels) and the frame rate is 30fps. After obtaining the video stream, a series of preprocessing operations are performed on each frame image:

[0040] First, the video sequence is de-burred to eliminate image blur caused by production line vibration or robot movement; then, histogram equalization and other algorithms can be used for light correction to reduce the impact of environmental light fluctuations on image brightness; further, the corrected color video frame is converted into a single-channel brightness image to simplify subsequent calculations. After the above operations, a series of injection molded part surface images available for analysis are obtained, denoted as a surface image set.

[0041] Finally, the first frame surface image in the surface image set is analyzed as an example: in order to efficiently analyze the local region, the first frame surface image is divided into multiple compact and similar feature sub-regions by using a superpixel segmentation algorithm, and these sub-regions are referred to as "analysis regions" in this embodiment.

[0042] S2, calculate the region noise degree based on the noise fluctuation in each analysis region; obtain the pixel noise confidence according to the deviation degree of the pixel gradient from the overall gradient feature of the analysis region where the pixel is located; combine the region noise degree, the pixel noise confidence and the gradient distribution to calculate the probability that each pixel belongs to noise.

[0043] S2.1, calculate the region noise degree based on the noise fluctuation in each analysis region.

[0044] After completing the region division, the noise fluctuation in each analysis region needs to be quantified. In industrial scenes, the manifestations of noise are diverse, which may be caused by highlights and shadows formed by uneven lighting, or by subtle surface material changes or dust. These factors together cause the instability of local brightness of the image. Therefore, a comprehensive index is needed to measure this instability. The present application constructs the region noise degree, which aims to jointly evaluate the visual "chaos" degree of a region from two dimensions of brightness fluctuation intensity and brightness range.

[0045] Accordingly, the region noise degree of the i-th analysis region in the j-th frame image is calculated as follows: The calculation formula of the region noise degree of the i-th analysis region in the j-th frame image is as follows:

[0046] ;

[0047] wherein denotes the region noise degree of the i-th analysis region in the j-th frame image; and are the variance and mean of the brightness values of all pixel points in the analysis region; and are the variance and mean of the brightness values of the pixels in the 8-neighborhood around the cluster center point of the analysis region; the above cluster center refers to the cluster center in each analysis region when the image is segmented into superpixels in step S1; and are the maximum and minimum values of the brightness in the analysis region; and are the maximum and minimum values of the brightness in the analysis region; denotes the normalization processing (such as maximum and minimum value normalization), which is used to map the result to the [0, 1] interval.

[0048] When any one or more of the overall brightness fluctuation (represented by ), the local brightness fluctuation of the core region (represented by ) and the brightness dynamic range (represented by ) in the analysis region increases, the sum result increases accordingly, finally leading to the increase of the region noise degree ​​The value also increases accordingly. In the invention scenario, this means that the system determines that the visual information in the area is more chaotic and disordered, that is, the area is more likely to be a high-reflectivity area or have other noise interference.

[0049] S2.2 Obtain the pixel noise confidence level based on the degree of deviation between the pixel gradient and the overall gradient features of the analysis area.

[0050] In step S2.1, a macroscopic noise score was given for each analysis region. However, this score may still fail to distinguish between two fundamentally different situations: one is random noise with chaotic gradient directions caused by specular reflection; the other is structural edges with relatively consistent gradient directions caused by real defects (such as scratches). Both may exhibit high brightness fluctuations. Therefore, the goal of this step is to delve into the pixel level and reveal the "intrinsic structure" of pixel changes by analyzing gradient information, thereby more accurately distinguishing between random noise and effective features.

[0051] The specific steps are as follows, still using the first... The first frame of the image Taking the first analysis region as an example: First, the Sobel operator is used to calculate the... Gradient magnitude of each pixel in the frame image and gradient direction For the first Each analysis region is used to calculate the average gradient magnitude of all pixels within that region. and the consistency coefficient of the gradient direction. Specifically, these include:

[0052] ;

[0053] in Indicates the first Consistency coefficient of the analysis regions; Indicates the first Pixels in a frame image The gradient direction; Indicates the first The total number of pixels within each analysis region. When all gradient directions are the same... Approaching 1, while in the case of complete randomness, The coefficient tends to 0, meaning that the smaller the consistency coefficient, the more dispersed the gradient direction is, and the greater the probability that it belongs to the highlight area; the larger the consistency coefficient, the more concentrated the gradient direction is, and the greater the probability that it belongs to the true edge.

[0054] Then in order to measure the deviation degree of single pixel point and the overall structural features of the region where it is located, the application constructs the concept of "pixel noise confidence". If the gradient direction and size of a pixel point are not consistent with the overall trend of the surrounding neighborhood, then it is likely to be an abnormal point (noise or defect). Accordingly, taking pixel point as an example, the calculation formula of its pixel noise confidence is constructed as follows:

[0055] ;

[0056] Among them represents the pixel noise confidence of pixel point ; represents the gradient direction of pixel point ; represents the vector mean of the gradient direction in the first analysis region; represents the gradient amplitude of pixel point ; represents the gradient amplitude mean of all pixel points in the first analysis region; is an adjustment coefficient, which can be set to 0.01 in the embodiment, used to adjust the relative gradient amplitude to avoid the case that the denominator is 0; represents the maximum gradient amplitude in the first frame image.

[0057] When the gradient direction of a certain pixel point deviates from the average gradient direction of the region where it is located, the value of the term is smaller, resulting in the increase of the value of the term . At the same time, if the gradient amplitude of the pixel point is much larger than the regional average value , the score term is also larger. The product of the two terms, i.e. the pixel noise confidence , will significantly increase. In the application scenario, a pixel point with a high value means that it is a "isolated point" or "mutation point" in structure, which is extremely inconsistent with the surrounding environment. Finally, the maximum and minimum normalization method can be used to record the normalized result as .

[0058] The reason for normalization is that different analysis regions may exhibit significant differences in the range of gradient magnitude and direction due to variations in lighting, material, or surface characteristics. Even if normalization is performed within a single region, the pixel noise confidence scores between different regions may still not be directly comparable. By performing intra-regional normalization on the pixel noise confidence scores within each analysis region, scale differences between different regions can be eliminated, making the pixel noise confidence scores of different regions comparable.

[0059] S2.3. Combine the regional noise level, pixel noise confidence level and gradient distribution to calculate the probability that each pixel belongs to noise.

[0060] At this point, information has been obtained from three key dimensions: regional macroscopic noise level ( ), pixel microstructure anomalies (normalized) ) and regional structural consistency ( Finally, a noise probability formula is constructed, which integrates the above information to arrive at a quantitative and more reliable judgment.

[0061] Also using pixels For example, the formula for calculating the probability that it belongs to noise is constructed as follows:

[0062] ;

[0063] in Represents pixels The true noise probability; Indicates the first The first frame of the image The regional noise level of each analysis area; Represents the normalized pixel points Pixel noise confidence; Indicates the first Consistency coefficient of the analysis regions.

[0064] Area noise level Normalized values ​​of pixel noise confidence Or the degree of dispersion of the gradient direction in the region Increasing any one of these terms will increase the sum value inside the sigmoid function, thus increasing the final noise probability. Increase it. Finally, a probability threshold can be set (for example, 0.6), and when the noise probability of a pixel is greater than the threshold, it is identified as a noise pixel.

[0065] S3. Adaptively adjust the smoothing intensity of the bilateral filter based on the actual noise probability.

[0066] After processing in step S2, the true noise probability is calculated for each pixel in the image. Traditional filters either smooth the image indiscriminately, potentially damaging details, or they are too conservative to remove persistent noise. Therefore, the goal of this step is to modify the bilateral filter using the obtained noise probability information, enabling it to perform differentiated processing on different pixels.

[0067] The specific operation is as follows: For pixels identified as noise, it is necessary to dynamically enhance their smoothing intensity. This invention achieves this by constructing a modified smoothing intensity factor.

[0068] The revised formula for calculating the smoothing intensity factor is as follows:

[0069] ;

[0070] in Represents pixels Corrected smoothing intensity factor; This represents the baseline smoothing intensity factor for traditional bilateral filtering; Represents pixels The true noise probability; This represents the intensity amplification factor, used to control the maximum increase in smoothing intensity. In this embodiment... It can be set to an experience value of 1; This represents a non-linear adjustment index, used to enhance the filtering effect on pixels with high noise probability. In this embodiment... It can be set to an experience value of 1.5.

[0071] When the noise probability of a pixel The higher the level, the more exponential the result. Sum of coefficients The amplification effect The larger this value, the greater the corrected smoothing strength factor. The larger the noise probability, the stronger the smoothing effect. In the invention scenario, this means that the bilateral filter applies a stronger smoothing effect when dealing with this high-noise-probability pixel, effectively "smoothing" it and blending it with surrounding pixels. Conversely, for pixels with low noise probability, Approaching 0 It is approximately equal to the benchmark value. The filtering effect is very gentle, thus preserving the true details of the image.

[0072] For pixels not identified as noise, the regular smoothing intensity factor is still used. The filtering processing is performed. In this way, the present application can effectively remove complex noises such as highlight areas while retaining the real edges and defect details of the surface of the injection molded part to the maximum, thereby significantly improving the quality of the monitoring image and laying a solid foundation for subsequent high-precision automatic defect detection.

[0073] At this point, the present application completes the adaptive denoising processing of each frame surface image in the injection molding production line monitoring video.

[0074] The injection molding production line real-time monitoring system based on machine vision embodiment:

[0075] On the other hand, the present application also provides an injection molding production line real-time monitoring system based on machine vision. As shown in Figure 2 The injection molding production line real-time monitoring system based on machine vision includes a processor and a memory, and the memory stores computer program instructions, which realize the injection molding production line real-time monitoring method based on machine vision according to the first aspect of the present application when executed by the processor.

[0076] The injection molding production line real-time monitoring system based on machine vision also includes other components such as communication interfaces that are well known to those skilled in the art, and their settings and functions are known in the art, so they will not be described here.

[0077] In the present application, the aforementioned memory can be any tangible medium containing or storing programs that can be used by or in conjunction with an instruction execution system, device or apparatus. For example, the computer readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory RRAM (Resistive Random Access Memory), dynamic random access memory DRAM (Dynamic Random Access Memory), static random access memory SRAM (Static Random-Access Memory), enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), high bandwidth memory HBM (High-Bandwidth Memory), hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store desired information and can be accessed by an application, module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device. Any application or module described in the present application can be implemented using computer readable / executable instructions stored or otherwise held by such computer readable medium.

Claims

1. Machine vision-based real-time monitoring method of injection molding production lines, characterized in that, The method comprises: acquiring a surface image of the injection molded part and segmenting the surface image into a plurality of analysis regions; calculating a region noise degree of each analysis region, the region noise degree being used to represent a degree of brightness fluctuation of the analysis region; calculating a pixel noise confidence of a pixel point in each analysis region, the pixel noise confidence being used to measure a deviation degree of a gradient of a single pixel point from a gradient feature of the analysis region as a whole to evaluate abnormality of a microstructure of the pixel point; obtaining a multi-dimensional information fusion noise probability model in combination with gradient direction consistency, region noise degree and pixel noise confidence of each analysis region, calculating a real noise probability of each pixel point through the noise probability model, and acquiring a corrected smoothing intensity factor based on the real noise probability, the corrected smoothing intensity factor being used to smooth noise in the image; adjusting smoothing intensity of each pixel point by a bilateral filtering algorithm according to the corrected smoothing intensity factor to obtain a denoised image.

2. The machine vision-based real-time monitoring method of injection molding production line according to claim 1, characterized in that, The calculation method of the real noise probability is specifically: ； in Represents pixels The true noise probability; Indicates the first The first frame of the image The regional noise level of each analysis area; Represents the normalized pixel points Pixel noise confidence; Indicates the first Consistency coefficient of the analysis regions.

3. The machine vision-based real-time monitoring method of injection molding production line according to claim 2, characterized in that, The calculation method of the pixel noise confidence is specifically: ； in Represents pixels Pixel noise confidence; Indicates the first Pixels in a frame image The gradient direction; Indicates the first The mean vector of gradient directions within each analysis region; Indicates the first Pixels in a frame image The gradient magnitude; Indicates the first The average gradient magnitude of all pixels within the analysis region; Indicates the adjustment coefficient; Indicates the first The maximum gradient magnitude in the frame image.

4. The machine vision-based real-time monitoring method of injection molding production line according to claim 2, characterized in that, The calculation method of the consistency coefficient is specifically: ； wherein represents the consistency coefficient of the th analysis region; represents the consistency coefficient of the th analysis region; represents the gradient direction of the pixel point in the th frame image; represents the total number of pixels in the th analysis region.

5. The machine vision-based real-time monitoring method of injection molding production line according to claim 2, characterized in that, The calculation method of the region noise degree is specifically: ； wherein represents the frame image, the region noise degree of the analysis region; and respectively represent the variance and the mean of the luminance values of all pixels in the analysis region; and respectively represent the variance and the mean of the luminance values of the pixels in the 8-neighborhood around the cluster center point of the analysis region; and respectively represent the maximum and the minimum of the luminance in the analysis region; represents a normalization function.

6. The machine vision-based real-time monitoring method of injection molding production lines according to claim 5, characterized in that, The analysis region clustering center point is obtained by a simple linear iterative clustering superpixel segmentation algorithm.

7. The machine vision-based real-time monitoring method of injection molding production lines according to claim 1, characterized in that, The calculation method of the corrected smoothing intensity factor is specifically: ； wherein represents the modified smoothing intensity factor of the pixel point ; represents the reference smoothing intensity factor of the traditional bilateral filter; represents the true noise probability of the pixel point ; represents the intensity amplification coefficient for controlling the maximum increment of the smoothing intensity; represents the non-linear adjustment index for strengthening the filtering effect on the pixel point with high noise probability.

8. The machine vision-based real-time monitoring method of injection molding production lines according to claim 7, characterized in that, For a noise point with a real noise probability less than a set threshold, a baseline smoothing intensity factor of the traditional bilateral filtering is used for smoothing processing.

9. The machine vision-based real-time monitoring method of injection molding production lines according to claim 1, characterized in that, The surface image of the injection molded part is acquired, comprising: acquiring a real-time video stream from an industrial camera deployed on an injection molding production line; performing preprocessing of de-bouncing and light correction on the video stream; converting the preprocessed video frame into a brightness image to obtain the surface image of the injection molded part.

10. Machine vision-based real-time monitoring system for injection molding production lines, characterized in that, The injection molding production line real-time monitoring method based on machine vision comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the injection molding production line real-time monitoring method based on machine vision.

Images