A multi-channel scrap chute number recognition method based on a crown block state perception

Abstract

The application discloses a kind of multi-channel scrap steel tank number identification method based on overhead crane state perception, it is related to scrap steel tank number identification technical field, the present application includes deployment camera and edge terminal, establish the data interface with overhead crane PLC, collect load and position data;According to load change detection hoist and match corresponding station passage;Analysis load difference feature discerns scrap steel tank bottom adhesion;Self-adapting acquisition single frame or multiple frames of images, carry out ROI cutting and target detection;When adhesion shakes, multiple frames of images are registered and recognized tank number based on gradient amplitude weighted fusion;The recognition result is reported to industrial internet platform.The present application can weaken motion blur influence, improve the recognition ability and data acquisition timeliness of scrap steel tank under bottom adhesion working condition by load transient feature discernment adhesion shakes, combined with multiple frames of gradient amplitude fusion.

Description

Technical Field

[0001] This invention belongs to the field of scrap steel trough number identification technology, and in particular relates to a multi-channel scrap steel trough number identification method based on crane status perception. Background Technology

[0002] In material transfer operations at scrap steel yards, scrap steel troughs are containers for loading scrap steel. Each trough's side wall is coated with a multi-digit number to record information such as the scrap steel's origin, batch, and weight. Overhead cranes, or bridge cranes, are the main equipment for lifting and transporting scrap steel troughs. They move along the factory's tracks and use lifting devices to grab the troughs and complete the transfer. Trough number identification refers to the automatic acquisition and entry of the numerical codes on the scrap steel troughs during overhead crane lifting, replacing manual transcription and improving data timeliness and accuracy.

[0003] To address the need for scrap steel trough number identification, existing technologies employ Radio Frequency Identification (RFID) and machine vision solutions. RFID solutions attach electronic tags to the surface of the scrap steel troughs, and readers positioned at workstations or overhead cranes read the numbers. However, in the high-temperature, strong magnetic field, and metal-shading environments of scrap steel yards, the electronic tags are susceptible to electromagnetic interference or physical damage, leading to reading failures. Machine vision solutions, on the other hand, use network cameras mounted above the yard to capture images of the scrap steel troughs, combining object detection and character recognition algorithms to automatically read the trough numbers. Some vision recognition systems also incorporate overhead crane position data to trigger image capture, allowing images to be taken when the crane reaches the workstation, reducing invalid captures during periods of no load.

[0004] However, scrap steel yards often have accumulated oil and water on the surface. In winter or low-temperature conditions, the bottom of the scrap steel troughs can become strongly adhered to the ground due to freezing or solidified oil. When using an overhead crane to lift the scrap steel troughs, the lifting equipment needs to pull the troughs off the ground. If there is bottom adhesion, the troughs will violently bounce and swing at the moment of lifting as they break free from the adhesion. This sudden shaking causes severe motion blur in the scrap steel trough images captured by the camera, with unclear character edges, making it difficult for subsequent trough number recognition algorithms to accurately extract numbers, resulting in frequent misidentification or recognition failures. Existing machine vision solutions generally rely on single-frame images for recognition, lacking effective means to identify and deal with the blur caused by lifting shaking. Under these conditions, the usability of recognition is significantly reduced, becoming the most prominent problem restricting the continuity of automatic data collection for scrap steel transfer. To address the above problems, the following solutions are proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-channel scrap steel trough number recognition method based on crane status perception. By monitoring the transient load changes during the crane lifting process in real time, the method identifies the adhesion state at the bottom of the scrap steel trough. When adhesion is detected, the method adaptively switches to multi-frame image acquisition and gradient amplitude-based fusion recognition. This method can preserve the edge of the trough number characters and suppress motion blur caused by shaking. It solves the problem that existing visual recognition schemes cannot accurately identify the trough number when the adhesion at the bottom of the scrap steel trough causes shaking during lifting due to blurring of a single frame image.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] This invention relates to a multi-channel scrap steel trough number identification method based on crane status perception, comprising the following steps:

[0008] Network cameras were deployed above multiple scrap steel trough workstations to establish a data interface between the edge computing terminal and the overhead crane PLC, and the mapping relationship between the workstation coordinate range and the camera channel was preset.

[0009] Real-time acquisition of video streams from each camera, as well as the load value and position coordinates of the overhead crane;

[0010] The load value is continuously monitored. When the load changes from zero to greater than zero and remains so, a lifting action is determined, and the corresponding camera channel is matched according to the position coordinates at the time of lifting.

[0011] Within a preset time window after the lifting action occurs, the load value sequence of the overhead crane is obtained, and based on the transient change characteristics of the load value sequence, it is identified whether the phenomenon of adhesion and breakage at the bottom of the scrap steel trough has occurred.

[0012] When no overlap occurs, a single frame is extracted from the video stream of the matching channel; when overlap occurs, multiple frames are extracted from the video stream of the matching channel.

[0013] The captured image is cropped with a ROI, and the scrap steel trough is detected using a target detection model. When the scrap steel trough is detected, the image region within the target bounding box is extracted.

[0014] When no adhesion occurs, digital recognition is performed on the image region within the target bounding box to obtain the scrap steel trough number string; when adhesion occurs, the image regions within the target bounding box of multiple frames are registered and fused, and digital recognition is performed on the fused image to obtain the scrap steel trough number string.

[0015] The identification results will be reported to the industrial internet platform.

[0016] Furthermore, the step of identifying whether the sticking and breaking off phenomenon at the bottom of the scrap steel trough has occurred based on the transient change characteristics of the load value sequence specifically includes:

[0017] Calculate the first difference of the load value sequence;

[0018] If there exists a first-order difference greater than a preset rising threshold, and within a preset number of sampling points after the difference greater than the rising threshold, there exists a first-order difference less than a preset falling threshold, then adhesion is determined to have occurred.

[0019] The preset rising threshold and the preset falling threshold are determined based on the rated lifting load of the overhead crane.

[0020] Furthermore, the preset rising threshold is the product of the rated lifting load and the first proportional coefficient, and the preset falling threshold is the product of the rated lifting load and the second proportional coefficient, wherein the first proportional coefficient is greater than the second proportional coefficient.

[0021] Furthermore, the capture of multiple frames refers to capturing K consecutive frames, where K is a preset integer greater than 1, and the K frames cover at least one main oscillation cycle of the crane's sway.

[0022] Furthermore, the registration and fusion processing of image regions within the target bounding boxes of multiple frames specifically includes:

[0023] Using the middle frame among the multiple frames as the reference frame, rigid body registration is performed on each of the remaining frames to estimate the translation and rotation angles between frames, and the estimated parameters are used to align the remaining frames with the reference frame.

[0024] Calculate the gradient magnitude for each aligned frame;

[0025] For each pixel location, the ratio of the gradient magnitude of that pixel in each frame to the sum of the gradient magnitudes of that pixel in all frames is used as the fusion weight. The pixel values ​​of the corresponding pixels in each frame are then summed in a weighted manner to obtain the fused image.

[0026] Furthermore, the rigid body registration employs a registration algorithm based on Fourier-Merlin transform.

[0027] Furthermore, the target detection model is the YOLO model, and the digit recognition model is the CRNN model.

[0028] Further, the step of performing digital recognition on the image region within the target bounding box to obtain the scrap steel trough number string, or performing digital recognition on the fused image to obtain the scrap steel trough number string, includes:

[0029] It identifies multiple characters and the confidence level of each character;

[0030] Sort the characters according to their X-axis coordinates within the target bounding box from smallest to largest;

[0031] Filter out characters with a confidence level lower than the preset confidence threshold, and concatenate the remaining characters in order to form the scrap steel trough number string.

[0032] Furthermore, the ROI cropping involves removing the surrounding background area from the image and retaining the effective image area containing the scrap steel trough, based on the preset pixel cropping parameters of the matched channel.

[0033] Furthermore, the identification results include the crane number, workstation number, timestamp, adhesion status identifier, and the scrap steel trough number string.

[0034] The present invention has the following beneficial effects:

[0035] This invention identifies the adhesion state at the bottom of scrap steel troughs by analyzing the transient load changes during overhead crane lifting. To address the swaying problem caused by adhesion and subsequent unwinding, it adaptively switches between multi-frame acquisition and fusion recognition strategies. Utilizing gradient amplitude weighting for multi-frame image fusion processing preserves character edges and suppresses motion blur, thereby improving the image quality of trough numbers under swaying conditions and enhancing the recognizability of blurred characters. Simultaneously, by combining the linkage triggering of overhead crane position and load, the recognition task for the corresponding channel is activated only after the lifting position is confirmed, reducing invalid image processing when there is no load and helping to reduce computational resource consumption. This method can maintain the continuous operation of the trough number recognition process under common working conditions such as oil contamination, rain, snow, and freezing in scrap steel yards, reducing manual re-entry due to recognition failures and improving the timeliness and completeness of data acquisition.

[0036] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating a multi-channel scrap steel trough number identification method based on crane status perception according to the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please see Figure 1 As shown, this invention is a multi-channel scrap steel trough number identification method based on crane status perception, comprising the following steps:

[0041] Network cameras were deployed above multiple scrap steel trough workstations to establish a data interface between the edge computing terminal and the overhead crane PLC, and the mapping relationship between the workstation coordinate range and the camera channel was preset.

[0042] Real-time acquisition of video streams from each camera, as well as the load value and position coordinates of the overhead crane;

[0043] Continuously monitor changes in load value. When the load changes from zero to greater than zero and remains at that value, determine that a lifting action has occurred, and match the corresponding camera channel based on the position coordinates at the time of lifting.

[0044] Within a preset time window after the lifting action occurs, the load value sequence of the overhead crane is obtained. Based on the transient change characteristics of the load value sequence, it is determined whether the phenomenon of adhesion and breakage at the bottom of the scrap steel trough has occurred.

[0045] When no overlap occurs, a single frame is extracted from the video stream of the matching channel; when overlap occurs, multiple frames are extracted from the video stream of the matching channel.

[0046] The captured image is cropped with a ROI, and the scrap steel trough is detected using a target detection model. When the scrap steel trough is detected, the image region within the target bounding box is extracted.

[0047] When no adhesion occurs, digital recognition is performed on the image region within the target bounding box to obtain the scrap steel trough number string; when adhesion occurs, the image regions within the target bounding boxes of multiple frames are registered and fused, and digital recognition is performed on the fused image to obtain the scrap steel trough number string.

[0048] The identification results will be reported to the industrial internet platform.

[0049] Based on the transient change characteristics of the load value sequence, it is possible to identify whether the scrap steel trough bottom adhesion and breakage phenomenon has occurred, specifically including:

[0050] Calculate the first difference of the load value sequence;

[0051] If there is a first-order difference greater than a preset rising threshold, and within a preset number of sampling points after the difference greater than the rising threshold, there is a first-order difference less than a preset falling threshold, then adhesion is determined to have occurred.

[0052] The preset rising threshold and preset falling threshold are determined based on the rated lifting load of the overhead crane.

[0053] The preset rising threshold is the product of the rated lifting load and the first proportional coefficient, and the preset falling threshold is the product of the rated lifting load and the second proportional coefficient, wherein the first proportional coefficient is greater than the second proportional coefficient.

[0054] Extracting multiple frames means extracting K consecutive frames, where K is a preset integer greater than 1, and the K frames cover at least one main oscillation cycle of the crane's sway.

[0055] The image regions within the target bounding boxes of multiple frames are registered and fused, specifically including:

[0056] Using the middle frame of a multi-frame set as the reference frame, rigid body registration is performed on each of the remaining frames to estimate the translation and rotation angles between frames, and the estimated parameters are used to align the remaining frames with the reference frame.

[0057] Calculate the gradient magnitude for each aligned frame;

[0058] For each pixel location, the ratio of the gradient magnitude of that pixel in each frame to the sum of the gradient magnitudes of that pixel in all frames is used as the fusion weight. The pixel values ​​of the corresponding pixels in each frame are then summed in a weighted manner to obtain the fused image.

[0059] Rigid body registration employs a registration algorithm based on Fourier-Merlin transform.

[0060] The target detection model is the YOLO model, and the digit recognition model is the CRNN model.

[0061] Digital recognition is performed on the image region within the target bounding box to obtain the scrap steel trough number string, or digital recognition is performed on the fused image to obtain the scrap steel trough number string, including:

[0062] It identifies multiple characters and the confidence level of each character;

[0063] Sort the characters according to their X-axis coordinates within the target bounding box from smallest to largest;

[0064] Filter out characters with a confidence level lower than the preset confidence threshold, and concatenate the remaining characters in sorted order to form the scrap steel trough number string.

[0065] ROI cropping removes the surrounding background area from the image based on the preset pixel cropping parameters of the matched channel, retaining the effective image area containing the scrap steel trough.

[0066] The identification results include the crane number, workstation number, timestamp, adhesion status indicator, and scrap steel trough number string.

[0067] The specific application of this embodiment is as follows:

[0068] To address the issue that scrap steel troughs in scrap steel yards freeze and stick to the ground due to oil stains, rain, and snow, causing violent shaking and blurry images during hoisting that make it impossible to accurately identify the trough number, the following complete implementation steps are proposed.

[0069] Step S1: System Deployment and Initialization

[0070] Above the scrap steel troughs No. 1, 2, and 3 in the scrap steel yard, network cameras were installed, respectively. Each camera outputs a 1080P video stream via the RTSP protocol. A YOLOv5-based scrap steel trough target detection model and a CRNN-based trough number recognition model were loaded into the edge computing terminal. Both models were trained using 1000 digital images of the scrap steel troughs from different angles and under different lighting conditions. An OPC UA data interface was established between the edge computing terminal and the overhead crane PLC to obtain the real-time load values ​​of overhead cranes No. 3 and No. 4. and X-axis position coordinates The mapping relationship between workstation areas and camera channels is preset in the edge computing terminal:

[0071] First workstation area: Corresponding to the first monitoring channel, denoted as ;

[0072] Second workstation area: The corresponding second monitoring channel is denoted as ;

[0073] Third workstation area: The corresponding third monitoring channel is denoted as .

[0074] Step S2: Real-time acquisition of multi-source data

[0075] The system starts two threads simultaneously. The video stream thread continuously pulls RTSP video streams from three network cameras, decodes each frame, and stores the latest frame into global variables. , , The data polling thread sends HTTP requests to the crane data server at a frequency of 1Hz to obtain the current load values ​​of cranes 3 and 4. It also records the timestamp of each frame. and X-axis position coordinates The returned JSON data is then parsed into a real-time status sequence, with each record in the sequence containing a timestamp. Crane number, load value and position coordinates .

[0076] Step S3: Lifting motion detection and adhesion status identification

[0077] Step S31: Lifting action detection. The edge computing terminal continuously monitors the load value. The change. If at two adjacent sampling points and Place, satisfy and And in the subsequent three consecutive sampling periods If the crane remains in this position, it is determined that a lifting action has occurred, and the lifting time is recorded as [time value missing]. .

[0078] Step S32: Workstation Matching and Channel Selection. Read... Position coordinates at time ,Will Compare with the three preset workstation intervals. If it falls within the first workstation interval, then select... If the passage falls within the second workstation area, select... If the passage falls within the third workstation area, select... If the channel does not fall into any interval, it is marked as "abnormal position" and the identification ends.

[0079] Step S33: Load transient feature extraction and adhesion state identification. At the moment of lifting... Then, based on the length of the time window Capture load value sequence in seconds , ,in , The data polling period is 1 second. Calculate the first difference for the load sequence:

[0080]

[0081] In the formula, For the first The change in the overhead crane load within each sampling interval; For the overhead crane at the sampling time The real-time load value is collected by the PLC; For the overhead crane at the previous sampling time Real-time load value; This is the sampling point index, with values ​​ranging from 1 to... Integers; This represents the total number of sampling points within the time window following the lifting moment. Based on the characteristic of a slow load increase during smooth lifting by the overhead crane in a non-adhesive state, a lifting threshold is set. and the drop threshold The method for obtaining the value is as follows:

[0082]

[0083] In the formula, The rated lifting load of the overhead crane is predetermined by the crane parameters. Traverse the difference sequence. If an index exists Make And in The following series Within each sampling point ( and At least one index exists. satisfy If this occurs, it is determined that the scrap steel at the bottom of the trough has stuck and broken free during the lifting process. In this case, the adhesion status indicator will be set. Set to 1; otherwise Set to 0.

[0084] Step S4: Adaptive Image Acquisition

[0085] Based on the channel number and adhesion status flag obtained in step S3 Implement differentiated image acquisition strategies.

[0086] like This indicates that the lifting is smooth and without adhesion, based on the corresponding global variable. The closest intercept time A single frame image, used as the image to be identified. .

[0087] like This indicates that adhesion has broken free and caused shaking, triggering multi-frame acquisition mode. From the video stream of the selected channel, Cache contiguous near time Frame images, respectively denoted as The time interval between adjacent frames is determined by the video frame rate to ensure coverage of at least one main oscillation cycle of the crane's sway, so as to capture relatively clear image information under different sway phases.

[0088] Step S5: ROI region cropping

[0089] Based on the preset pixel cropping parameters of the selected channel, the single frame image is processed. Or multiple frames per frame ROI processing was performed to remove the surrounding material yard background area, retaining only the effective image area containing the scrap steel trough, resulting in the cropped image. or .

[0090] Step S6: Target detection in the scrap steel trough

[0091] The cropped image is fed into the scrap steel trough target detection model. If the model does not detect any scrap steel trough targets, a "detection anomaly" flag is directly generated, and the process jumps to step S8 to report the anomaly information. If a scrap steel trough is detected, the target bounding box coordinates and the image region within the bounding box are output. For multi-frame fusion scenarios, the intermediate frame is used. Based on the detection results, sub-images within the corresponding bounding boxes of each frame are extracted uniformly. .

[0092] Step S7: Slot Number Identification

[0093] Step S71: Routine identification during smooth lifting. When At that time, directly use the target bounding box image The data is fed into a digit recognition model to obtain a string of candidate characters and their confidence levels.

[0094] Step S72: Multi-frame fusion recognition during adhesion and shaking. When At that time, for the extracted Sub-images Perform the following blending operation to eliminate motion blur and enhance character edges.

[0095] First, take the intermediate frame As the reference frame, for each of the remaining frames A rigid body registration algorithm based on Fourier-Merlin transform is used to estimate the inter-frame translation. and rotation angle Using estimated parameters Align with the reference frame to obtain the registered image. Then, each frame is calculated. In pixels gradient magnitude at :

[0096]

[0097] In the formula, For the first Image after frame registration In pixel coordinates The gradient magnitude at a point reflects the edge strength at that point; For the first The image obtained after registering (translation and rotation alignment) the frame sub-images; For image In the horizontal direction ( The first-order partial derivative of (direction) is calculated by Sobel level operator convolution; For image In the vertical direction ( The first-order partial derivative of the direction is calculated by convolution using the Sobel vertical operator; These represent the row and column coordinates of the image pixels, respectively. Taking into account the specific feature that the scrap steel trough numbers are mostly horizontal and vertical strokes with strong edge contrast, a fusion weight based on gradient magnitude is constructed:

[0098]

[0099] In the formula, For the first Frame image in pixels The fusion weight coefficient at the point ranges from 0 to 1. For the first After frame registration, the image is in Gradient magnitude at; This represents the total number of frames buffered in multi-frame acquisition mode. ; The summation variable is used to iterate through frames 1 to 1. frame; To prevent small constants with denominators of zero, take Final fused image It is given by the following formula:

[0100]

[0101] In the formula, For the final image after fusion, in pixels The grayscale value (or channel value) at that location will be used for slot number identification; For the calculated first Frame in The fusion weight at the location; For the first After frame registration, the image is in Pixel value at; Frame sequence number index, from 1 to ( ); Total number of frames;

[0102] After merging, the sharp edges of the character areas are preserved, while the blurred areas caused by shaking are automatically suppressed. The data is fed into a digit recognition model to obtain a set of characters and their confidence levels.

[0103] Step S73: Post-processing of recognition results. Regardless of whether conventional or fusion recognition is used, the recognized characters are sorted in ascending order of their X-axis coordinates within the bounding box, and noisy characters with a confidence level below 0.6 are filtered out. The remaining characters are then concatenated into a complete string of scrap steel trough numbers.

[0104] Step S8: Reporting the recognition results

[0105] The edge computing terminal constructs a JSON format message, including the crane number, workstation number, timestamp, and connection status. The scrap steel trough number string or abnormal identifier is sent to the data interface of the industrial internet platform via an HTTP POST request to complete a complete scrap steel trough number identification and recording.

[0106] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0107] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-channel scrap steel trough number identification method based on crane status perception, characterized in that, The identification method includes the following steps: Network cameras were deployed above multiple scrap steel trough workstations to establish a data interface between the edge computing terminal and the overhead crane PLC, and the mapping relationship between the workstation coordinate range and the camera channel was preset. Real-time acquisition of video streams from each camera, as well as the load value and position coordinates of the overhead crane; The load value is continuously monitored. When the load changes from zero to greater than zero and remains so, a lifting action is determined, and the corresponding camera channel is matched according to the position coordinates at the time of lifting. Within a preset time window after the lifting action occurs, the load value sequence of the overhead crane is obtained, and based on the transient change characteristics of the load value sequence, it is identified whether the phenomenon of adhesion and breakage at the bottom of the scrap steel trough has occurred. When no overlap occurs, a single frame is extracted from the video stream of the matching channel; when overlap occurs, multiple frames are extracted from the video stream of the matching channel. The captured image is cropped with a ROI, and the scrap steel trough is detected using a target detection model. When the scrap steel trough is detected, the image region within the target bounding box is extracted. When no adhesion occurs, digital recognition is performed on the image region within the target bounding box to obtain the scrap steel trough number string; when adhesion occurs, the image regions within the target bounding box of multiple frames are registered and fused, and digital recognition is performed on the fused image to obtain the scrap steel trough number string. The identification results will be reported to the industrial internet platform.

2. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The method of identifying whether the sticking and breaking off phenomenon at the bottom of the scrap steel trough has occurred based on the transient change characteristics of the load value sequence specifically includes: Calculate the first difference of the load value sequence; If there exists a first-order difference greater than a preset rising threshold, and within a preset number of sampling points after the difference greater than the rising threshold, there exists a first-order difference less than a preset falling threshold, then adhesion is determined to have occurred. The preset rising threshold and the preset falling threshold are determined based on the rated lifting load of the overhead crane.

3. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 2, characterized in that, The preset rising threshold is the product of the rated lifting load and the first proportional coefficient, and the preset falling threshold is the product of the rated lifting load and the second proportional coefficient, wherein the first proportional coefficient is greater than the second proportional coefficient.

4. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The process of capturing multiple frames involves capturing K consecutive frames, where K is a preset integer greater than 1, and the K frames cover at least one main oscillation cycle of the crane's sway.

5. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The registration and fusion process for image regions within the target bounding boxes of multiple frames specifically includes: Using the middle frame among the multiple frames as the reference frame, rigid body registration is performed on each of the remaining frames to estimate the translation and rotation angles between frames, and the estimated parameters are used to align the remaining frames with the reference frame. Calculate the gradient magnitude for each aligned frame; For each pixel location, the ratio of the gradient magnitude of that pixel in each frame to the sum of the gradient magnitudes of that pixel in all frames is used as the fusion weight. The pixel values ​​of the corresponding pixels in each frame are then summed in a weighted manner to obtain the fused image.

6. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 5, characterized in that, The rigid body registration adopts a registration algorithm based on Fourier-Merlin transform.

7. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The target detection model is the YOLO model, and the digit recognition model is the CRNN model.

8. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The step of performing digital recognition on the image region within the target bounding box to obtain the scrap steel trough number string, or performing digital recognition on the fused image to obtain the scrap steel trough number string, includes: It identifies multiple characters and the confidence level of each character; Sort the characters according to their X-axis coordinates within the target bounding box from smallest to largest; Filter out characters with a confidence level lower than the preset confidence threshold, and concatenate the remaining characters in order to form the scrap steel trough number string.

9. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The ROI cropping is performed by removing the surrounding background area in the image and retaining the effective image area containing the scrap steel trough, based on the preset pixel cropping parameters of the matched channel.

10. The multi-channel scrap steel trough number identification method based on crane status perception according to claim 1, characterized in that, The identification results include the crane number, workstation number, timestamp, adhesion status identifier, and the waste steel trough number string.

Images