Online detection method for LED light bar

Abstract

The invention discloses an on-line detection method for an LED light bar. The method comprises the following steps: acquiring a surface image of the LED light bar in real time through an industrial camera and synchronously recording the movement speed; performing dynamic compensation processing on the image to eliminate motion blur; balancing brightness and suppressing interference through a self-adaptive gamma correction algorithm; detecting defects and outputting type and position information by using a multi-scale feature fusion network; and triggering the rejecting device to mark or sort according to the defect position. According to the method, the image quality is improved through dynamic compensation and brightness balance, the defects such as micron-sized cracks and abnormal welding spots are accurately recognized in combination with multi-scale feature fusion, meanwhile, sorting delay is corrected in real time based on motion parameters, the problems of missing detection, false detection and positioning deviation in a high-speed production scene are solved, the detection precision reaches the submillimeter level, the false detection rate is reduced, and the production efficiency is improved. And the production yield and efficiency are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of LED light bar detection, and in particular to an online detection method for LED light bar. Background Art

[0002] LED light strips, as highly efficient and energy-efficient lighting components, are widely used in display devices, decorative lighting, and industrial lighting. Their production involves multiple steps, including packaging, welding, and die bonding. Even the slightest defect in any step, such as a cold solder joint, a broken LED, or a short circuit, can render the finished product inoperable. To ensure product quality, online inspection technology has become an essential component of the production process.

[0003] Currently, quality inspection of LED light strips primarily relies on manual visual inspection and offline automated testing. Manual visual inspection suffers from low efficiency, poor consistency, and fatigue, making it difficult to meet the demands of large-scale production. Within automated inspection technologies, machine vision-based inspection systems are becoming increasingly mainstream. These systems use industrial cameras to capture images of light strips and combine them with image processing algorithms to identify defects. However, existing technologies still face significant bottlenecks in practical application. Due to dynamic jitter, uneven brightness, and ambient light interference during the high-speed, continuous production of LED light strips, traditional image processing algorithms are susceptible to noise during feature extraction, resulting in reduced accuracy in identifying subtle defects such as micron-level solder cracks and localized brightness anomalies. For example, under high-speed conditions, existing systems may miss defects due to insufficient image acquisition frame rates or lagging algorithm processing speeds. Furthermore, due to halo interference from highly reflective light strip surfaces or adjacent LEDs, traditional threshold segmentation methods struggle to accurately distinguish between normal light spots and abnormal areas, further increasing the risk of misjudgment. These issues are particularly prominent in the inspection of flexible or miniaturized, high-density light strips, severely limiting the reliability of inspection systems and production yield. Therefore, there is an urgent need to develop an online detection method that can adapt to high-speed continuous production environments, has anti-interference capabilities, and can accurately identify subtle defects to solve the above problems. Summary of the Invention

[0004] The present invention provides an online detection method for LED light strips, so as to solve the problem that the existing LED light strip detection method has a decreased recognition accuracy rate for subtle defects during high-speed continuous production.

[0005] The present invention provides an online detection method for LED light strips, comprising the following steps:

[0006] Step S1: using an industrial camera to capture the surface image of the LED light bar in real time, and synchronously record the movement speed during image acquisition;

[0007] Step S2: performing dynamic compensation processing on the surface image to eliminate motion blur in the image based on the motion speed;

[0008] Step S3: Perform brightness equalization processing on the compensated image, and suppress brightness unevenness and ambient light interference through an adaptive gamma correction algorithm;

[0009] Step S4: performing defect detection on the processed image based on a multi-scale feature fusion network, and outputting defect type and location information;

[0010] Step S5: Generate a real-time positioning signal based on the defect position information, and trigger a rejection device to mark or sort the defective lamp beads.

[0011] Furthermore, performing dynamic compensation processing on the surface image to eliminate motion blur in the image based on the motion speed includes:

[0012] Obtain the real-time motion parameters of the LED light bar through the encoder, including linear speed and acceleration;

[0013] Building a dynamic compensation model based on the motion parameters and restoring high-frequency details of the blurred area through a deconvolution algorithm;

[0014] Adjust image acquisition parameters based on the restored high-frequency details to optimize the exposure time and frame rate of subsequent frames.

[0015] Furthermore, the compensated image is brightness-equalized, and an adaptive gamma correction algorithm is used to suppress uneven brightness and ambient light interference, including:

[0016] Divide the compensated image into multiple sub-regions and calculate the grayscale histogram of each sub-region respectively;

[0017] Local contrast limitation is applied to high-brightness sub-regions, and nonlinear stretching is performed on low-brightness sub-regions.

[0018] The processed sub-regions are merged into an image with global brightness balance through a weighted fusion algorithm.

[0019] Furthermore, the processed sub-regions are merged into a globally brightness balanced image through a weighted fusion algorithm, including:

[0020] Calculate the fusion weight based on the grayscale gradient difference of adjacent sub-regions;

[0021] Bilinear interpolation algorithm is used to eliminate brightness jumps at the sub-region boundaries;

[0022] The color consistency of the fused image is verified through multi-channel color space conversion.

[0023] Furthermore, the image acquisition parameters are adjusted based on the restored high-frequency details to optimize the exposure time and frame rate of subsequent frames, including:

[0024] Generate an exposure time adjustment coefficient based on the high-frequency detail distribution of the previous frame;

[0025] Predict the blur intensity of the next frame based on the acceleration change rate and dynamically adjust the trigger frequency of the industrial camera;

[0026] The image acquisition parameters are iteratively optimized through a feedback controller until the dynamic compensation error is lower than a set threshold.

[0027] Furthermore, defect detection is performed on the processed image based on a multi-scale feature fusion network, and defect type and location information is output, including:

[0028] Extract local microscopic features from images through convolutional neural networks;

[0029] Extract global context features through the Transformer module;

[0030] The local micro features are fused with the global context features using channel-wise attention weighting to generate a defect probability map.

[0031] Furthermore, the local micro features are fused with the global context features using channel attention weighted fusion to generate a defect probability map, including:

[0032] Perform spatial pyramid pooling on local micro features to generate multi-resolution feature vectors;

[0033] Filter global context features related to defects through a gating mechanism;

[0034] The Softmax function is used to classify the weighted fusion features and output the defect type confidence.

[0035] Furthermore, a real-time positioning signal is generated based on the defect location information, and a rejection device is triggered to mark or sort the defective lamp beads, including:

[0036] Mapping the defect location information to the physical coordinate system of the LED light strip;

[0037] Calculate the delay time for defective lamp beads to reach the rejection device based on the movement speed;

[0038] The pulse signal triggers the pneumatic nozzle to mark the defective lamp beads with ink after a set delay.

[0039] Furthermore, the delay time for the defective lamp beads to reach the rejection device is calculated according to the movement speed, including:

[0040] Generate time compensation according to the movement speed and mechanical response delay of the rejection device;

[0041] Use Kalman filter to predict the position deviation of LED light bar within the delay time;

[0042] The delay time is dynamically corrected in combination with the position deviation to ensure that the mark position error is less than 0.1 mm.

[0043] Furthermore, in step S1, the acquisition frame rate of the industrial camera and the movement speed of the LED light bar satisfy the following relationship:

[0044] Frame rate ≥ (motion speed / minimum detectable defect size) × safety factor, where the safety factor is 1.2~2.0.

[0045] The beneficial effects of the present invention include: It eliminates image blur caused by high-speed motion through dynamic compensation processing, and combines an adaptive brightness equalization algorithm to suppress ambient light interference, significantly improving image quality; utilizing a multi-scale feature fusion network to simultaneously extract local microscopic features and global contextual features, effectively solving the problems of missed detection and misjudgment of subtle defects; and ensuring accurate sorting of defective lamp beads through real-time positioning signals and a delay correction mechanism. This method can achieve real-time detection and closed-loop control of LED light strip defects in high-speed continuous production scenarios, achieving sub-millimeter detection accuracy, significantly reducing the false detection rate, and significantly improving production yield and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0047] Figure 1 The present invention provides a flow chart of an online detection method for LED light strips. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. The technical solutions provided by each embodiment of the present invention are described in detail below in conjunction with the drawings.

[0049] See also Figure 1 , an embodiment of the present invention provides an online detection method for LED light strips, comprising the following steps:

[0050] Step S1: The surface image of the LED light strip is collected in real time by an industrial camera, and the movement speed during image collection is synchronously recorded.

[0051] The industrial camera captures real-time surface images of the LED light strip during its continuous motion, while simultaneously recording the light strip's speed using an encoder or sensor. This simultaneous recording of speed provides key parameters for subsequent dynamic compensation, ensuring that image processing matches the light strip's actual motion. Precise speed synchronization avoids mismatches between image and motion parameters caused by acquisition delays, laying the foundation for eliminating motion blur. The industrial camera's acquisition frame rate and the LED light strip's speed satisfy the following relationship:

[0052] Frame rate ≥ (motion speed / minimum detectable defect size) × safety factor, where the safety factor is 1.2~2.0.

[0053] Step S2: performing motion compensation processing on the surface image to eliminate motion blur in the image based on the motion speed.

[0054] In this embodiment, step S2 includes: obtaining the real-time motion parameters of the LED light bar, including linear velocity and acceleration, through an encoder; constructing a dynamic compensation model based on the motion parameters, and restoring the high-frequency details of the blurred area using a deconvolution algorithm; adjusting image acquisition parameters based on the restored high-frequency details to optimize the exposure time and frame rate of subsequent frames. Specifically, the following steps are performed: generating an exposure time adjustment coefficient based on the distribution of high-frequency details in the previous frame; dynamically adjusting the trigger frequency of the industrial camera by predicting the blur intensity of the next frame based on the rate of change of acceleration; and iteratively optimizing the image acquisition parameters using a feedback controller until the dynamic compensation error falls below a set threshold.

[0055] Motion compensation uses a deconvolution algorithm combined with motion speed parameters to restore high-frequency details in areas blurred by high-speed motion. This step uses motion speed to construct a blur kernel model, inferring the image degradation caused by motion, and ultimately restoring a clear image. Motion compensation significantly improves image resolution, preserving the details of micron-level defects such as cracked solder joints and broken LEDs, thus avoiding missed detections due to blur.

[0056] Step S3: Perform brightness equalization processing on the compensated image, and suppress brightness unevenness and ambient light interference through an adaptive gamma correction algorithm.

[0057] In this embodiment, step S3 includes: dividing the compensated image into multiple subregions and calculating the grayscale histogram of each subregion; applying local contrast constraints to high-brightness subregions and nonlinear stretching to low-brightness subregions; and merging the processed subregions into a globally brightness-balanced image using a weighted fusion algorithm. Specifically, fusion weights are calculated based on the grayscale gradient differences between adjacent subregions; a bilinear interpolation algorithm is used to eliminate brightness jumps at subregion boundaries; and color consistency of the fused image is verified through multi-channel color space conversion.

[0058] Brightness equalization divides the image into multiple sub-regions, applies local contrast limitation to high-brightness areas, and nonlinear stretching to low-brightness areas. Finally, a weighted fusion algorithm is used to merge the images into a globally brightness-balanced image. Adaptive gamma correction dynamically adjusts parameters based on the grayscale distribution of each sub-region, suppressing brightness fluctuations caused by variations in LED light levels or ambient light interference. This ensures a consistent brightness baseline for the input image to the defect detection algorithm, reducing the risk of misjudgment.

[0059] Step S4: Perform defect detection on the processed image based on the multi-scale feature fusion network and output defect type and location information.

[0060] In this embodiment, step S4 includes: extracting local micro features in the image through a convolutional neural network; extracting global context features through a Transformer module; performing channel attention weighted fusion of local micro features and global context features to generate a defect probability map, specifically: performing spatial pyramid pooling on local micro features to generate multi-resolution feature vectors; screening global context features related to defects through a gating mechanism; and classifying the weighted fused features using a Softmax function to output the defect type confidence.

[0061] The multi-scale feature fusion network uses a convolutional neural network to extract local microscopic features, such as solder joint shape and LED surface texture. It also uses a Transformer module to extract global contextual features, such as LED arrangement patterns and circuit connectivity. These features are weighted and fused using a channel-wise attention mechanism to generate a defect probability map, accurately distinguishing between normal and defective areas. This network can simultaneously capture the local features of subtle defects and overall structural anomalies, addressing the problem of missed detections caused by traditional single-scale feature extraction.

[0062] Step S5: Generate a real-time positioning signal based on the defect position information, and trigger a rejection device to mark or sort the defective lamp beads.

[0063] In this embodiment, step S5 includes mapping the defect location information to the physical coordinate system of the LED light bar; calculating the delay time for the defective LED to reach the rejection device based on the movement speed. Specifically, this involves generating a time compensation based on the movement speed and the mechanical response delay of the rejection device; predicting the position deviation of the LED light bar during the delay time using a Kalman filter; and dynamically correcting the delay time based on the position deviation to ensure that the marking position error is less than 0.1mm. A pulse signal triggers the pneumatic nozzle to inkjet mark the defective LED after the set delay.

[0064] By mapping defect location information to the physical coordinate system of the LED light bar and combining it with real-time motion speed to calculate the delay before the defective LED reaches the rejection device, a pulse signal is used to trigger a pneumatic nozzle or robotic arm for precise marking or sorting. This step dynamically corrects the delay time, compensating for mechanical response delays and motion position deviations, ensuring that the sorting action is fully synchronized with the defect location and avoiding positioning errors caused by high-speed motion.

[0065] As can be seen from the above embodiments, the present invention solves the problem of detail loss caused by image blur in high-speed motion scenes through real-time synchronization of dynamic compensation processing and motion parameters; adopts a partitioned brightness equalization algorithm to effectively suppress the influence of the LED light bar's own luminous differences and ambient light interference on the detection accuracy; through a multi-scale feature fusion network, combined with the complementarity of local microscopic features and global contextual information, the ability to identify subtle defects is significantly improved; further, through real-time positioning signals and delay correction mechanisms, the accurate sorting of defective lamp beads in high-speed continuous production scenarios is achieved. The synergistic effect of the above technical features enables the present invention to reduce the false detection rate to below 0.5% while maintaining high detection efficiency, and the detection accuracy reaches sub-millimeter level, which is suitable for the online quality control needs of high-density, miniaturized LED light bars.

[0066] An embodiment of the present invention further provides a storage medium storing a computer program that, when executed by a processor, implements some or all of the steps of each embodiment of the online detection method for LED light strips provided by the present invention. The storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0067] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software and a necessary general-purpose hardware platform. Based on this understanding, the technical solutions in the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, or an optical disk, and includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute the methods described in various embodiments of the present invention, or portions thereof.

[0068] The above-described embodiments of the present invention do not limit the protection scope of the present invention.

Claims

1. An online detection method for LED light strips, characterized in that: The following steps are involved: Step S1: using an industrial camera to capture the surface image of the LED light bar in real time, and synchronously record the movement speed during image acquisition; Step S2: performing dynamic compensation processing on the surface image to eliminate motion blur in the image based on the motion speed; Step S3: Perform brightness equalization processing on the compensated image, and suppress brightness unevenness and ambient light interference through an adaptive gamma correction algorithm; Step S4: performing defect detection on the processed image based on a multi-scale feature fusion network, and outputting defect type and location information; Step S5: Generate a real-time positioning signal based on the defect position information, and trigger a rejection device to mark or sort the defective lamp beads.

2. The online detection method for LED light strips according to claim 1, wherein: Performing dynamic compensation processing on the surface image to eliminate motion blur in the image based on the motion speed, including: Obtain the real-time motion parameters of the LED light bar through the encoder, including linear speed and acceleration; Building a dynamic compensation model based on the motion parameters and restoring high-frequency details of the blurred area through a deconvolution algorithm; Adjust image acquisition parameters based on the restored high-frequency details to optimize the exposure time and frame rate of subsequent frames.

3. The online detection method for LED light strips according to claim 1, wherein: Perform brightness equalization on the compensated image, and suppress uneven brightness and ambient light interference through an adaptive gamma correction algorithm, including: Divide the compensated image into multiple sub-regions and calculate the grayscale histogram of each sub-region respectively; Local contrast limitation is applied to high-brightness sub-regions, and nonlinear stretching is performed on low-brightness sub-regions. The processed sub-regions are merged into an image with global brightness balance through a weighted fusion algorithm.

4. The online detection method for LED light strips according to claim 3, wherein: The processed sub-regions are merged into an image with global brightness balance through a weighted fusion algorithm, including: Calculate the fusion weight based on the grayscale gradient difference of adjacent sub-regions; Bilinear interpolation algorithm is used to eliminate brightness jumps at the sub-region boundaries; The color consistency of the fused image is verified through multi-channel color space conversion.

5. The online detection method for LED light strips according to claim 2, wherein: Adjust image acquisition parameters based on the restored high-frequency details to optimize the exposure time and frame rate of subsequent frames, including: Generate an exposure time adjustment coefficient based on the high-frequency detail distribution of the previous frame; Predict the blur intensity of the next frame based on the acceleration change rate and dynamically adjust the trigger frequency of the industrial camera; The image acquisition parameters are iteratively optimized through a feedback controller until the dynamic compensation error is lower than a set threshold.

6. The online detection method for LED light strips according to claim 1, wherein: Defect detection is performed on the processed image based on a multi-scale feature fusion network, and defect type and location information is output, including: Extract local microscopic features from images through convolutional neural networks; Extract global context features through the Transformer module; The local micro features are fused with the global context features using channel-wise attention weighting to generate a defect probability map.

7. The online detection method for LED light strips according to claim 6, characterized in that: The local micro features are fused with the global context features using channel-wise attention weighting to generate a defect probability map, including: Perform spatial pyramid pooling on local micro features to generate multi-resolution feature vectors; Filter global context features related to defects through a gating mechanism; The Softmax function is used to classify the weighted fusion features and output the defect type confidence.

8. The online detection method for LED light strips according to claim 1, wherein: Generate real-time positioning signals based on defect location information and trigger the rejection device to mark or sort defective lamp beads, including: Mapping the defect location information to the physical coordinate system of the LED light strip; Calculate the delay time for defective lamp beads to reach the rejection device based on the movement speed; The pulse signal triggers the pneumatic nozzle to mark the defective lamp beads with ink after a set delay.

9. The online detection method for LED light strips according to claim 8, characterized in that: Calculate the delay time for defective lamp beads to reach the rejection device based on the movement speed, including: Generate time compensation according to the movement speed and mechanical response delay of the rejection device; Use Kalman filter to predict the position deviation of LED light bar within the delay time; The delay time is dynamically corrected in combination with the position deviation to ensure that the mark position error is less than 0.1 mm.

10. The online detection method for LED light strip according to claim 1, characterized in that: In step S1, the acquisition frame rate of the industrial camera and the movement speed of the LED light bar satisfy the following relationship: Frame rate ≥ (motion speed / minimum detectable defect size) × safety factor, where the safety factor is 1.2~2.0.

Images