Image identification and laser scanning fused smelting slag ladle hook identification method

Abstract

The invention provides a smelting slag ladle hook identification method fusing image identification and laser scanning, and belongs to the technical field of intelligent manufacturing. The method comprises the steps that bimodal images of a smelting slag ladle hook and a trunnion are collected and preprocessed, and an enhanced bimodal image is obtained; based on the enhanced bimodal image, utilizing an improved YOLOv5 model to obtain a bimodal image recognition result; based on the bimodal image recognition result, a control instruction is sent out, and the control instruction is manual operation when the smelting slag ladle hook is not hung into the trunnion; when the smelting slag ladle hook is hung into the trunnion, the control instruction is that a laser scanning recognition result is obtained through laser scanning recognition, and a dispatching instruction is sent out according to the laser scanning recognition result; according to the method, the robustness of smelting slag ladle hook identification is improved through a double guarantee mechanism of matching a visible light-infrared bimodal fusion identification technology with laser point cloud verification, the accuracy of smelting slag ladle hook identification in various environments is ensured, and the construction safety is ensured.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and particularly to a recognition method for the hanging hook of a smelting slag ladle that integrates image recognition and laser scanning. Background Art

[0002] The transfer of smelting slag ladles in metallurgical enterprises is a crucial link in the production process. It not only affects the continuity of the production process but also the safety of the entire operation. As an important container for loading and transferring, the smelting slag ladle plays a vital role in production. Currently, in metallurgical enterprises, the hanging operation of the slag ladle hook and trunnion relies on on-site confirmation by operators. This not only requires operators to have high professional skills and rich experience but also may lead to serious safety hazards in case of operation errors.

[0003] The prior art "Method for realizing the confirmation of ladle trunnion hanging by using video imaging technology" (CN117612068A): The system implemented in this patent installs a thermal imaging and visible light dual-spectrum camera on each side of the ladle trunnion, and uses the video data obtained by the two cameras for deep learning algorithm recognition. By comparing the analysis results of the two cameras, the accuracy and reliability of the recognition of the hanging state of the smelting slag ladle hook are judged. However, the image recognition method is too dependent on the quality of the collected images, resulting in insufficient detection accuracy.

[0004] Therefore, a recognition method for the hanging hook of a smelting slag ladle that integrates image recognition and laser scanning is needed. Summary of the Invention

[0005] In view of this, the present invention provides a recognition method for the hanging hook of a smelting slag ladle that integrates image recognition and laser scanning, which combines dual-mode image recognition and laser scanning recognition to accurately detect the hanging state of the trunnion.

[0006] For this purpose, the present invention provides the following technical solutions:

[0007] A recognition method for the hanging hook of a smelting slag ladle that integrates image recognition and laser scanning, comprising:

[0008] Collecting dual-mode images of the hanging hook and trunnion of the smelting slag ladle;

[0009] Preprocessing the dual-mode images of the hanging hook and trunnion of the smelting slag ladle to obtain enhanced dual-mode images;

[0010] Based on the enhanced dual-mode images, using an improved YOLOv5 model to obtain dual-mode image recognition results;

[0011] Based on the dual-mode image recognition results, issuing a control instruction;

[0012] The issuing of the control instruction based on the dual-mode image recognition results includes:

[0013] When the bimodal image recognition result is that the hook of the smelting slag ladle is not hooked into the trunnion, the control instruction is to switch to manual operation;

[0014] When the bimodal image recognition result is that the hook of the smelting slag ladle is hooked into the trunnion, the control instruction is to use laser scanning recognition to obtain the laser scanning recognition result; and issue a scheduling instruction according to the laser scanning recognition result;

[0015] The bimodal image includes: an infrared image and a visible light image.

[0016] Further, the issuing of the scheduling instruction according to the laser scanning recognition result includes:

[0017] When the laser scanning recognition result is that the hook of the smelting slag ladle is hooked into the trunnion, this recognition ends;

[0018] When the laser scanning recognition result is that the hook of the smelting slag ladle is not hooked into the trunnion, the scheduling instruction is to switch to manual operation.

[0019] Further, the obtaining of the laser scanning recognition result through laser scanning recognition includes:

[0020] Obtain the point cloud data of the hook of the smelting slag ladle and the trunnion through laser scanning;

[0021] Perform adaptive processing and feature extraction on the point cloud data of the hook of the smelting slag ladle and the trunnion to obtain the key feature points of the trunnion axis center and the hook center of the smelting slag ladle hook;

[0022] Use the key feature points of the trunnion axis center and the hook center of the smelting slag ladle hook for rough registration, and execute the improved iterative closest point algorithm with a fusion adaptive strategy to output the registered point cloud;

[0023] Locate the three-dimensional coordinates of the trunnion axis center and the hook center of the smelting slag ladle hook based on the registered point cloud;

[0024] Obtain the distance between the trunnion axis center and the hook center of the smelting slag ladle hook through the three-dimensional coordinates of the trunnion axis center and the hook center of the smelting slag ladle hook;

[0025] When the distance from the hook center of the smelting slag ladle to the trunnion axis center is within the safety threshold, the laser scanning recognition result is that the trunnion has been hooked into the smelting slag ladle hook;

[0026] When the distance from the hook center of the smelting slag ladle to the trunnion axis center is not within the safety threshold, the laser scanning recognition result is that it is determined that the trunnion is not hooked into the smelting slag ladle hook.

[0027] Further, the obtaining of the bimodal image recognition result by using the improved YOLOv5 model based on the enhanced bimodal image includes:

[0028] Feature extraction is performed through the backbone network CSPDarknet53 to obtain multi-scale image features;

[0029] According to the real-time environmental light data and temperature data, an adaptive weight coefficient is calculated through a preset environmental parameter mapping function to perform weight allocation for visible light image features and infrared image features;

[0030] The coordinate prediction values of the centroid of the smelting slag ladle hook and the centroid of the trunnion are respectively output through the detection head;

[0031] Based on the coordinate prediction values of the centroid of the smelting slag ladle hook and the centroid of the trunnion, the actual distance between the centroid of the smelting slag ladle hook and the trunnion is calculated;

[0032] When the actual distance between the centroid of the smelting slag ladle hook and the trunnion is greater than the preset distance threshold, the bimodal image recognition result is that the smelting slag ladle hook is not hung on the trunnion;

[0033] When the actual distance between the centroid of the smelting slag ladle hook and the trunnion is less than or equal to the preset distance threshold, the bimodal image recognition result is that the smelting slag ladle hook has been hung on the trunnion.

[0034] Further, the improved iterative closest point algorithm with a fusion adaptive strategy includes:

[0035] Based on the consistency of point cloud curvature and normal vector, adaptive weight allocation is performed for edge, corner region and flat region;

[0036] Based on the registration error threshold, the step size is dynamically adjusted.

[0037] Further, the improved YOLOv5 model includes:

[0038] An SE attention mechanism is embedded in each residual module of the backbone network CSPDarknet53.

[0039] Further, the preprocessing of the bimodal image of the smelting slag ladle hook and the trunnion includes:

[0040] Performing registration alignment on the bimodal image of the smelting slag ladle hook and the trunnion;

[0041] Performing light compensation on the bimodal image of the smelting slag ladle hook and the trunnion;

[0042] Performing smoothing and denoising on the bimodal image of the smelting slag ladle hook and the trunnion.

[0043] Advantages and positive effects of the present invention:

[0044] The present invention uses a dual - guarantee mechanism that combines visible - light and infrared dual - mode fusion recognition technology with laser point cloud verification, enabling the system to reach a relatively high level of reliability; improving the robustness of the recognition of the smelting slag ladle hook, ensuring the accuracy of the recognition of the smelting slag ladle hook in various environments, and guaranteeing construction safety.

[0045] The visible - light and infrared dual - mode fusion recognition technology of the present invention adopts the YOLOv5 recognition model combined with the SE attention mechanism to optimize feature extraction, effectively reducing the misrecognition rate; and in the daytime strong - light environment, by reducing the visible - light weight and increasing the infrared weight, it effectively avoids glare interference; in the nighttime low - light condition, by increasing the infrared weight, it ensures a stable recognition effect and guarantees all - weather recognition accuracy.

[0046] The laser scanning recognition technology of the present invention uses an adaptive noise reduction algorithm to significantly improve the quality of point cloud data; uses an improved iterative closest point algorithm for registration to achieve high - precision position judgment; FPFH feature extraction ensures a stable key - point matching effect; overall, it improves the accuracy of laser recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0048] Figure 1 It is the flowchart of the method in the embodiment of the present invention;

[0049] Figure 2 It is the recognition flowchart of the improved YOLOv5 model in the embodiment of the present invention;

[0050] Figure 3 It is the laser scanning recognition flowchart in the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] In order to enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0052] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0053] The present invention provides a method for identifying a smelting slag ladle hook that combines image recognition and laser scanning. The system scheduling terminal synchronously calls a visible light camera and an infrared camera for bimodal image acquisition; the acquired bimodal image recognition module performs fusion analysis on the acquired images to judge the state of the smelting slag ladle hook. If the bimodal image recognition result is that the smelting slag ladle hook has been hung on the trunnion, laser scanning recognition is used to judge the state of the smelting slag ladle hook again. If the bimodal image recognition result is that the smelting slag ladle hook has not been hung on the trunnion, the scheduling terminal will issue an instruction to request manual intervention. The point cloud data obtained by laser scanning recognition is processed and analyzed, and the laser scanning recognition result is fed back to the scheduling terminal: if the laser scanning recognition result is that the smelting slag ladle hook has been hung on the trunnion, the scheduling terminal will judge that the trunnion state is hung in. If the laser scanning recognition result is that the smelting slag ladle hook has not been hung on the trunnion, the scheduling terminal will issue an instruction to request manual intervention.

[0054] Combined with Figure 2 Further explanation of image recognition in the present invention;

[0055] Step 1: When there is a task of identifying the hanging of the smelting slag ladle hook during the production process, the scheduling terminal synchronously calls a visible light camera and an infrared camera to perform bimodal image acquisition of the smelting slag ladle hook and the trunnion part.

[0056] Step 2: Preprocess the acquired bimodal images of the smelting slag ladle hook and the trunnion, including:

[0057] 1) Perform image registration and alignment. In this embodiment, the visible light image and the infrared image are respectively converted into grayscale images to reduce the data volume and simplify the processing.

[0058] Normalize the visible light and infrared grayscale images to unify the pixel value range of the images;

[0059] Calculate the histograms of the two images respectively to obtain the distribution of each pixel intensity. Calculate the joint histogram of the bimodal images to evaluate the statistical dependence between the images.

[0060] Based on the joint histogram, calculate the mutual information value. Measure the amount of information shared between two images through mutual information. The larger the mutual information, the higher the degree of alignment between the images. Map one image to the coordinate system of another image through a translation transformation model. Adjust the transformation parameters through an optimization algorithm to maximize the mutual information value. When the mutual information value converges or reaches the preset number of iterations, stop the optimization. Transform one image according to the optimized transformation parameters, including: such as translation, rotation or affine transformation, to align it with another image. In this embodiment, the optimization algorithm is the gradient descent method.

[0061] 2) Because the change of illumination conditions will affect the brightness, contrast and color distribution of the image, resulting in the degradation of image quality.

[0062] In this embodiment, through the illumination compensation algorithm, by adjusting the histogram distribution of the bimodal image respectively, the brightness and contrast of the image are made more uniform.

[0063] 3) Perform noise filtering processing. Preferably, use a Gaussian kernel to smooth the image.

[0064] Step 3: Obtain the bimodal image recognition result through the improved YOLOv5 deep learning model based on the preprocessed bimodal images of the smelting slag ladle hook and trunnion.

[0065] 1) In this embodiment, the backbone network adopts CSPDarknet53; input the processed bimodal image into the backbone network CSPDarknet53, reduce redundant calculations through the cross-stage local connection structure, and at the same time use multi-scale convolutional layers to extract image features at different levels. To further optimize the feature expression ability, embed the SE attention mechanism in each residual module of the backbone network, and strengthen the response of the key feature channels and suppress the interference of irrelevant backgrounds through dynamic calibration of channel weights.

[0066] 2) According to the real-time environmental light intensity and temperature data, calculate the adaptive weight coefficient through the preset environmental parameter mapping function to enhance the infrared feature weight when the light is weak and enhance the visible light texture feature in a high-temperature environment.

[0067] 3) Use a feature pyramid network for multi-scale feature fusion;

[0068] 3) Identify the centroid of the smelting slag ladle hook and the centroid of the trunnion through the detection head.

[0069] Adopt a decoupled detection structure to output the coordinate predictions of the centroid of the smelting slag ladle hook and the centroid of the trunnion respectively, and further obtain the distance between the centroid of the smelting slag ladle hook and the centroid of the trunnion.

[0070] Step 4: Obtain the dual - mode image recognition result based on the distance value between the centroid of the smelting slag ladle hook and the centroid of the trunnion and a preset threshold value:

[0071] When the distance value between the centroid of the smelting slag ladle hook and the centroid of the trunnion is less than the preset threshold value, the dual - mode image recognition result is that the smelting slag ladle hook has been hung on the trunnion;

[0072] When the distance value between the centroid of the smelting slag ladle hook and the centroid of the trunnion is greater than or equal to the preset threshold value, the dual - mode image recognition result is that the smelting slag ladle hook has not been hung on the trunnion;

[0073] Step 5: Feed back the dual - mode image recognition result to the dispatching end, and the dispatching end issues control instructions according to the dual - mode image recognition result, including:

[0074] When the dual - mode image recognition result is that the smelting slag ladle hook has been hung on the trunnion, perform laser scanning recognition;

[0075] When the dual - mode image recognition result is that the smelting slag ladle hook has not been hung on the trunnion, perform manual intervention.

[0076] The laser scanning recognition process is as Figure 3 shown and includes:

[0077] Step 1: Multi - source point cloud data acquisition and pre - processing;

[0078] Obtain the three - dimensional point cloud data of the trunnion axis and the hook center of the smelting slag ladle through a lidar. For the problems of point cloud sparsity and noise caused by dust and high temperature in the smelting scenario, perform adaptive processing on the point cloud data:

[0079] 1) Calculate the local point cloud density based on the k - nearest neighbor algorithm, and dynamically divide the high - density area and the low - density area; in this embodiment, the high - density area is the trunnion and the hook surface of the smelting slag ladle, and the low - density area is the edge part.

[0080] 2) Set different noise reduction thresholds according to the density distribution. The high - density area uses a strict threshold to filter out fine noise, and the low - density area relaxes the threshold to avoid over - filtering;

[0081] In this embodiment, the high - density area uses a neighborhood distance standard deviation < 0.05m for filtering, and the low - density area has a standard deviation < 0.1m;

[0082] 3) Apply a statistical filter to remove outliers that deviate from the local point cloud mean by more than a preset multiple of the standard deviation, and retain the key structural features.

[0083] Step 2: Multi - level geometric feature extraction and key point screening;

[0084] Perform geometric feature analysis on the noise - reduced point cloud:

[0085] 1) Calculate the normal vector of the point cloud based on the PCA algorithm to characterize the local surface orientation;

[0086] 2) Construct a Simplified Point Feature Histogram (SPFH) through parameters such as the normal vector angle and distance to describe the geometric properties of the single-point neighborhood;

[0087] 3) FPFH feature enhancement: Combine the SPFH features of the current point and its k-neighborhood points, and perform weighted fusion with the reciprocal of the distance as the weight to generate enhanced FPFH features, improving the characterization ability of local details such as the trunnion arc surface and the hook groove of the smelting slag ladle;

[0088] 4) Key point screening: Calculate the variance of the FPFH (Fast Point Feature Histograms) feature vectors of all points, set a dynamic variance threshold, and screen out the points in the regions with significant curvature changes (trunnion edge, tip of the hook of the smelting slag ladle) as key feature points;

[0089] 5) Feature matching: Based on the FPFH feature similarity (cosine distance < 0.2), establish initial matching point pairs between the trunnion axis and the hook center point cloud of the smelting slag ladle, providing a constraint relationship for rough registration.

[0090] Step 3: On the basis of rough registration, execute the improved iterative closest point algorithm with a fusion adaptive strategy to refine the alignment of the point cloud. The specific process is as follows:

[0091] 1) Optimize feature matching: Combine the FPFH feature and the ISS (Intrinsic Shape Signatures) feature descriptor to extract highly repeatable feature points, and eliminate the mismatched point pairs through the RANSAC algorithm (Random Sample Consensus algorithm), retaining the geometrically consistent matches;

[0092] 2) Adaptive weight assignment: Based on the point cloud curvature and normal vector consistency, assign a weight of 1.0 to the edge and corner regions and a weight of 0.3 to the flat regions to strengthen the alignment of key structures; Pay more attention to the regions with significant features and improve the registration accuracy.

[0093] Dynamic step size adjustment: The initial iteration uses a large search step size: translation 0.1m, rotation 5°, to quickly approach the global optimum. When the registration error reduction rate is lower than 10%, switch to a small step size: translation 0.01m, rotation 0.5°, for fine optimization; Improve the algorithm efficiency.

[0094] 3) Weighted least squares optimization: Construct an objective function to minimize the weighted sum of the squares of the distances between the matching point pairs.

[0095] 4) Update the rigid body transformation matrix according to the optimization results: the rotation matrix and the translation vector, and iterate cyclically until convergence; in this embodiment, the convergence condition is that the error change amount is less than the threshold or the maximum number of iterations is reached.

[0096] 5) Output the registered fused point cloud and the final transformation matrix, providing accurate spatial alignment data for subsequent key point extraction.

[0097] Step 5: Based on the registered point cloud, locate the three-dimensional coordinates of the trunnion axis center and the hook center of the smelting slag ladle hook respectively:

[0098] 1) Extraction of the trunnion axis center:

[0099] Extract the point cloud in the trunnion area through geometric filtering; use the least squares method to fit the cylinder model, solve the parameters of the central axis of the cylinder, and take the midpoint of the axis as the trunnion axis center coordinate.

[0100] 2) Extraction of the hook center of the smelting slag ladle hook:

[0101] Extract the inner contour point set of the hook ring of the smelting slag ladle hook based on curvature analysis; use the least squares method to fit a three-dimensional space circle, and combine the normal vector constraint to ensure that the circle plane is consistent with the force direction of the smelting slag ladle hook, and calculate the center coordinate as the hook center coordinate of the smelting slag ladle hook.

[0102] Step 6: Obtain the laser scanning recognition result through the trunnion axis center coordinate and the hook center coordinate of the smelting slag ladle hook;

[0103] Calculate the distance from the hook center coordinate of the smelting slag ladle hook to the trunnion axis center coordinate and set a safety threshold;

[0104] When the distance from the hook center of the smelting slag ladle hook to the trunnion axis center is within the safety threshold, the laser scanning recognition result is that the trunnion has been hung into the hook of the smelting slag ladle;

[0105] When the distance from the hook center of the smelting slag ladle hook to the trunnion axis center is not within the safety threshold, the laser scanning recognition result is that it is determined that the trunnion has not been hung into the hook of the smelting slag ladle;

[0106] Step 7: The dispatching end issues a control instruction according to the laser scanning recognition result;

[0107] When the laser scanning recognition result is that the hook of the smelting slag ladle has been hung into the trunnion, the recognition is completed.

[0108] When the laser scanning recognition result is that the hook of the smelting slag ladle has not been hung into the trunnion, an instruction for manual intervention is issued.

[0109] The present invention combines the visible light-infrared dual-modal fusion recognition technology with the lidar scanning technology, further improving the accuracy of identifying the hook of the smelting slag ladle in various environments, improving the detection quality, and ensuring construction safety.

[0110] Among them, the visible light-infrared dual-modal fusion recognition technology has all-weather and all-time recognition capabilities. In the daytime strong light environment, by reducing the weight of visible light and increasing the weight of infrared, glare interference is effectively avoided; in the night weak light condition, by increasing the weight of infrared, a stable recognition effect is ensured; therefore, the recognition accuracy of single visible light recognition is significantly improved.

[0111] Moreover, the combination of the improved deep learning algorithm and the multi-source data fusion technology significantly improves the recognition accuracy and reliability; specifically, the Feature Pyramid Network realizes multi-scale detection, greatly improving the detection accuracy of targets at different scales; the SE attention mechanism optimizes feature extraction, effectively reducing the misrecognition rate; the dual-modal data fusion combined with the multiple guarantee mechanism of laser point cloud verification enables the system to reach a high reliability level.

[0112] For the laser scanning recognition stage, the adaptive noise reduction algorithm significantly improves the quality of point cloud data; the improved Iterative Closest Point (ICP) algorithm realizes high-precision position judgment; the FPFH feature extraction ensures a stable key point matching effect; combined with the adaptive compensation algorithm, the system can work stably within a wide temperature range.

[0113] Using the method of the present invention effectively replaces manual visual inspection, significantly reducing the operation risk of personnel; and the detection process takes a short time, improving the production efficiency; after the system is put into use, the incidence rate of related safety accidents is greatly reduced; and it effectively reduces the labor cost and significantly improves the production efficiency.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A recognition method for the hanging hook of a smelting slag ladle that integrates image recognition and laser scanning, characterized in that, Including: Collecting dual-modal images of the hanger and trunnion of the smelting slag ladle; Preprocessing the dual-modal images of the hanger and trunnion of the smelting slag ladle to obtain enhanced dual-modal images; Based on the enhanced dual-modal images, using an improved YOLOv5 model to obtain dual-modal image recognition results; Based on the dual-modal image recognition results, issuing control instructions; The issuing of control instructions based on the dual-modal image recognition results includes: When the dual-modal image recognition result is that the hanger of the smelting slag ladle is not hung on the trunnion, the control instruction is to switch to manual operation; When the dual-modal image recognition result is that the hanger of the smelting slag ladle is hung on the trunnion, the control instruction is to use laser scanning recognition to obtain laser scanning recognition results; and according to the laser scanning recognition results, issuing scheduling instructions; The dual-modal images include: infrared images and visible light images.

2. The method for identifying the hanging hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 1, characterized in that The issuing of scheduling instructions according to the laser scanning recognition results includes: When the laser scanning recognition result is that the hanger of the smelting slag ladle is hung on the trunnion, the current recognition ends; When the laser scanning recognition result is that the hanger of the smelting slag ladle is not hung on the trunnion, the scheduling instruction is to switch to manual operation.

3. The method for identifying the hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 1, wherein The obtaining of laser scanning recognition results through laser scanning recognition includes: Obtaining point cloud data of the hanger and trunnion of the smelting slag ladle through laser scanning; Performing adaptive processing and feature extraction on the point cloud data of the hanger and trunnion of the smelting slag ladle to obtain key feature points of the trunnion axis center and the hook center of the hanger of the smelting slag ladle; Using the key feature points of the trunnion axis center and the hook center of the hanger of the smelting slag ladle for rough registration, and performing an improved iterative closest point algorithm with a fusion adaptive strategy to output registered point clouds; Based on the registered point clouds, locating the three-dimensional coordinates of the trunnion axis center and the hook center of the hanger of the smelting slag ladle; Obtaining the distance between the trunnion axis center and the hook center of the hanger of the smelting slag ladle through the three-dimensional coordinates of the trunnion axis center and the hook center of the hanger of the smelting slag ladle; When the distance from the hook center of the hanger of the smelting slag ladle to the trunnion axis center is within the safety threshold, the laser scanning recognition result is that the trunnion has been hung on the hanger of the smelting slag ladle; When the distance from the hook center of the hanger of the smelting slag ladle to the trunnion axis center is not within the safety threshold, the laser scanning recognition result is that it is judged that the trunnion has not been hung on the hanger of the smelting slag ladle.

4. The method for identifying the hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 1, characterized in that, The obtaining of dual-modal image recognition results based on the enhanced dual-modal images using an improved YOLOv5 model includes: Performing feature extraction through the backbone network CSPDarknet53 to obtain multi-scale image features; According to the real-time environmental light data and temperature data, calculating an adaptive weight coefficient through a preset environmental parameter mapping function to perform weight allocation for visible light image features and infrared image features; Respectively outputting coordinate prediction values of the centroid of the hanger of the smelting slag ladle and the centroid of the trunnion through the detection head; Calculating the actual distance between the centroid of the hanger of the smelting slag ladle and the centroid of the trunnion based on the coordinate prediction values of the centroid of the hanger of the smelting slag ladle and the centroid of the trunnion; When the actual distance between the centroid of the hanger of the smelting slag ladle and the centroid of the trunnion is greater than the preset distance threshold, the dual-modal image recognition result is that the hanger of the smelting slag ladle is not hung on the trunnion; When the actual distance between the centroid of the hanger of the smelting slag ladle and the centroid of the trunnion is less than or equal to the preset distance threshold, the dual-modal image recognition result is that the hanger of the smelting slag ladle is hung on the trunnion.

5. The method for identifying the hanging hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 3, wherein, The improved iterative closest point algorithm with a fusion adaptive strategy includes: Adaptive weight distribution is performed for the edge, corner region, and flat region based on the consistency of point cloud curvature and normal vector; The step size is dynamically adjusted based on the registration error threshold.

6. The method for identifying the hanging hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 4, wherein The improved YOLOv5 model includes: Embedding the SE attention mechanism in each residual module of the backbone network CSPDarknet53.

7. The method for identifying the hook of the smelting slag ladle integrating image recognition and laser scanning according to claim 1, wherein The preprocessing of the dual-modal images of the smelting slag ladle hook and trunnion includes: Registering and aligning the dual-modal images of the smelting slag ladle hook and trunnion; Performing illumination compensation on the dual-modal images of the smelting slag ladle hook and trunnion; Performing smoothing and denoising on the dual-modal images of the smelting slag ladle hook and trunnion.

Images