Production workshop danger source early warning system based on image recognition technology

Abstract

The invention belongs to the technical field of image recognition, and particularly relates to a production workshop hazard source early warning system based on an image recognition technology, a data acquisition module is used for deploying a high-definition camera, a mobile acquisition device and an environment sensor; the data annotation and management module is used for annotating the acquired images; the algorithm training and classification module is used for training an SSD algorithm according to the image data so as to output a classification recognition result; and the early warning module is used for comparing the alarm threshold values of the corresponding target categories according to the output result, and generating alarm mechanisms of different levels. The data annotation and management module adopts an image annotation tool based on an OpenCV software library to carry out automatic annotation on the acquired image; and the sample training and target detection efficiency of a subsequent SSD algorithm is improved. The data processing module extracts the target area of the dangerous source image, consumption of computing resources by irrelevant backgrounds is reduced, and higher detection precision and computing efficiency are achieved for the SSD algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to a production workshop hazard source early warning system based on image recognition technology. Background Art

[0002] In modern production environments, monitoring and management of workshop hazards is an important part of ensuring production safety and improving efficiency. However, traditional hazard identification methods face many shortcomings:

[0003] 1. Traditional safety management mainly relies on manual monitoring of the workshop environment by patrol personnel, which cannot achieve all-weather, high-precision real-time hazard warning. Especially when facing complex production environments or hidden danger sources, this method is prone to missed reports and false alarms.

[0004] Second: Traditional solutions are mostly based on a single-dimensional (image or sensor) data source, making it difficult to simultaneously analyze multi-dimensional data on dangerous sources. This detection method, which lacks multimodal data support, limits the system's ability to comprehensively judge dangerous scenarios and makes it difficult to provide comprehensive and accurate early warning data.

[0005] Third: The recognition method based on traditional machine learning is limited by the sample quality and the performance of the algorithm model, and has the problem of insufficient detection accuracy, especially when facing multiple targets and complex backgrounds, which further limits the optimization space of the model. Summary of the invention

[0006] In order to solve the technical problems that traditional solutions lack a way to simultaneously analyze multi-dimensional data of hazardous sources; and traditional machine learning is limited by insufficient sample quality and algorithm accuracy, the present invention provides a production workshop hazardous source early warning system based on image recognition technology.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] The production workshop hazard source early warning system based on image recognition technology includes data collection module, data annotation and management module, detection model training module and early warning module;

[0009] The data acquisition module is used to deploy high-definition cameras, mobile acquisition devices, and environmental sensors; high-definition cameras are used to collect real-time images of the workshop and provide original images for algorithm training and detection;

[0010] The mobile collection device includes an aircraft or a robot vehicle equipped with a camera; the environmental sensor includes an infrared thermometer, a smoke sensor or a humidity sensor;

[0011] The environmental sensor is configured on the mobile collection device and follows the mobile collection device to detect environmental data.

[0012] The data labeling and management module is used to label the collected images in COCO format, define the basic information, target category, bounding box coordinates and environmental data of the workshop hazard source images (based on OpenCV's image labeling tool LabelImg) and store them in a structured directory in JSON file format for classification management; (COCO labeling format is recommended, supporting more complex labeling structures and multi-target labeling)

[0013] Preferably, the basic information of the workshop hazard source image includes the image file name, image size and unique identifier;

[0014] Target categories include overheated electrical appliances, shelf structure displacement, and personnel regulations; unique identifiers are used to uniquely label image IDs and target categories;

[0015] The bounding box coordinates are pixel value coordinates;

[0016] Environmental data include target temperature, humidity, smoke concentration and shelf structure displacement data.

[0017] Preferably, the method further comprises: posting a positioning label frame on the boundary of the specific physical object in the marked image to implement the boundary frame positioning management of the physical object;

[0018] Preferably, the method further includes: performing data enhancement (such as rotation, flipping, etc.) on the annotated image to increase the diversity of input samples;

[0019] Preferably, the method further includes: screening verified workshop hazard source images in the OpenCV library and annotating them accordingly as samples of input images, thereby expanding the sample diversity of the input images.

[0020] Optionally, a data processing module is also included before the algorithm training and classification module, which is used to extract the target area of ​​the acquired image; reduce redundant data to optimize computing resources; and crop the extracted target region (ROI) and convert it into the image input format required by the SSD algorithm.

[0021] Preferably, the specific process of extracting the target area from the collected image in the data processing module includes:

[0022] Use OpenCV's edge detection algorithm (such as Canny) to parse the bounding box coordinates according to the annotation file and locate the target region (ROI) of the image.

[0023] The target area of ​​the image is further binarized to extract the pixel data of the target area.

[0024] Preferably, the method further comprises: utilizing OpenCV's super-resolution reconstruction technology to enlarge the target area and enhance the image, so as to improve the detection model's ability to recognize and analyze the target.

[0025] The algorithm training and classification module is used to train the SSD algorithm based on image data to output classification recognition results;

[0026] Preferably, the specific process of the algorithm training and classification module includes:

[0027] Data preparation: Use COCO format annotation files to load training data;

[0028] SSD algorithm construction: select the basic network of the SSD algorithm (such as VGG-16 or MobileNet); adjust the number of target categories and default box configuration of the output layer according to the scene requirements; use the classification loss function and the positioning loss function as the loss function; use SGD or Adam as the optimizer to optimize the algorithm parameters;

[0029] Write a custom data loader (such as PyTorch’s Dataset class) to read the input image.

[0030] Algorithm verification and optimization: Evaluate the algorithm performance on the validation set and calculate the mean average precision (mean AveragePrecision); adjust hyperparameters (such as learning rate, loss function, optimizer) to optimize the algorithm;

[0031] Algorithm deployment: The newly acquired images are optimized by the data processing module and then input into the trained SSD algorithm, which outputs the image’s hazardous source target category, bounding box coordinates, and environmental data.

[0032] The early warning module is used to generate different levels of warning mechanisms (such as sound and light alarms, SMS notifications) based on the output results and the alarm thresholds of the corresponding target categories.

[0033] Preferably, the specific process of the early warning module includes:

[0034] Classification management is carried out according to the target category of the image hazard source, specifically:

[0035] When the temperature of the appliance exceeds 60°C, a second-level alarm is triggered; when the temperature of the appliance exceeds 80°C, a first-level alarm is triggered;

[0036] When the shelf structure displacement exceeds 3cm, a second-level alarm is triggered; when the shelf structure displacement exceeds 5cm, a first-level alarm is triggered;

[0037] When a person enters the workshop without wearing safety equipment, a level 2 alarm is triggered; when a person enters the workshop without wearing safety equipment and appears in a high-risk area, a level 1 alarm is triggered;

[0038] When the smoke concentration is detected to be greater than 100ppm, a level 1 alarm is triggered.

[0039] Beneficial effects of the present invention:

[0040] 1. The production workshop hazard source early warning system of the present invention is based on image recognition technology, uses high-definition cameras and mobile devices to collect images in all directions, and combines multimodal data such as environmental sensor data (temperature, humidity, smoke concentration), thereby improving the accuracy of hazard source identification and the effective data basis for judgment.

[0041] 2. The data annotation and management module based on the OpenCV software library uses image annotation tools to set annotation modes, specifications and file formats to automatically annotate the collected images, thereby improving the sample training and target detection efficiency of the subsequent SSD algorithm. The data processing module extracts the target area of ​​the hazard source image, reduces the consumption of computing resources by irrelevant background, and provides efficient and formatted input data for the SSD detection algorithm.

[0042] 3. Image annotation and target area extraction help the SSD algorithm achieve higher detection accuracy and computational efficiency. Based on the output results of the algorithm, a multi-level early warning mechanism is set to quickly respond to different levels of dangerous sources and realize intelligent workshop safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0044] Figure 1 This is a structural framework diagram of the production workshop hazard source early warning system based on image recognition technology of the present invention;

[0045] Figure 2 This is a flow chart of the algorithm training and classification module in the production workshop hazard source early warning system based on image recognition technology of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] See also Figure 1-Figure 2 As shown, the production workshop hazard source early warning system based on image recognition technology includes a data acquisition module, a data annotation and management module, a detection model training module and an early warning module;

[0048] The data acquisition module is used to deploy high-definition cameras, mobile acquisition devices, and environmental sensors; high-definition cameras are used to collect real-time images of the workshop and provide original images for algorithm training and detection;

[0049] The mobile collection device includes an aircraft or a robot vehicle equipped with a camera; the environmental sensor includes an infrared thermometer, a smoke sensor or a humidity sensor;

[0050] The environmental sensor is configured on the mobile collection device and follows the mobile collection device to detect environmental data.

[0051] Specifically, the data acquisition module is the basis of the entire production workshop hazard source early warning system. Its function is to obtain multi-source and multi-dimensional workshop hazard source images and environmental data, and provide data support for subsequent image annotation, detection model training and identification.

[0052] During the specific implementation process, high-definition cameras are fixedly deployed in key areas of the workshop (such as equipment-intensive areas, storage areas, areas with frequent personnel activities) and areas with high incidence of hazardous sources to collect high-resolution real-time images.

[0053] Mobile data acquisition equipment can specifically be aircraft (drones) or robotic vehicles with cameras: images collected on high-rise shelves or dense areas that are difficult for cameras to cover are used to make up for the blind spots of fixed cameras. At the same time, mobile data acquisition equipment is equipped with high-definition cameras, infrared thermometers and other sensors to collect images and environmental data along preset trajectories or autonomous navigation paths. Among them, infrared thermometers are used to obtain abnormal temperature on the surface of equipment or in local areas. Smoke sensors detect possible precursors to fire (such as increased smoke concentration). Humidity sensors monitor the humidity of the workshop environment or water stains on the ground to identify risks that affect the safety of equipment or transportation of goods. Improve the accuracy of hazard identification and the effective data basis for judgment.

[0054] In addition, it should be noted that the collected images and environmental data are uploaded to the storage system of the data annotation and management module through the transmission module (such as a wireless network or a wired connection). And ensure the temporal consistency of data collection: through timestamp recording, maintain the correlation between the image and the environmental data.

[0055] The data labeling and management module is used to label the collected images in COCO format, define the basic information, target category, bounding box coordinates and environmental data of the workshop hazard source images (based on OpenCV's image labeling tool LabelImg) and store them in a structured directory in JSON file format for classification management; (COCO labeling format supports more complex labeling structures and multi-target labeling)

[0056] Furthermore, the basic information of the workshop hazard source image includes the image file name, image size, and unique identifier;

[0057] Target categories include overheated electrical appliances, shelf structure displacement, and personnel regulations; unique identifiers are used to uniquely label image IDs and target categories;

[0058] The bounding box coordinates are pixel value coordinates;

[0059] Environmental data include target temperature, humidity, smoke concentration and shelf structure displacement data.

[0060] Furthermore, it also includes: posting a positioning label frame on the boundary of the specific physical object in the marked image to achieve boundary frame positioning management of the physical object;

[0061] Specifically, the data annotation and management module is a key link in the production workshop hazard source early warning system based on image recognition technology. By annotating the collected workshop hazard source images, it can provide high-quality data input for algorithm training and real-time recognition.

[0062] The data annotation and management module of the present invention uses open source image annotation tools such as LabelImg of OpenCV or VGG Image Annotator (VIA), and automatically annotates the collected images by setting the annotation mode, specification and file format.

[0063] Among them, the annotation mode improves the model's recognition ability for complex scenes by defining the basic information of the workshop hazard source image, target category, bounding box coordinates and environmental data.

[0064] The basic information of the image includes: 'file_name': the image file name. 'width' and 'height': the image dimensions. 'id': the unique identifier of the image.

[0065] and 'category_id': target category id, including overheating of electrical appliances, displacement of shelf structure, and personnel regulations. For more detailed classification, please refer to the open source database of OpenCV to filter the original image samples.

[0066] 'bbox = [x, y, width, height]': The bounding box coordinates are pixel value coordinates; usually represented by four values: the upper left corner coordinates (x, y) and the lower right corner coordinates (x, y) or the center point coordinates (x, y, width, height). The bounding box coordinates define the location range of the target in the image, making it easier for the algorithm to focus on the target area.

[0067] Environmental data including target temperature, humidity, smoke concentration and shelf structure displacement data are embedded in the annotation file as additional attributes to form a complete data set.

[0068] The annotation specification uses the COCO format and supports more complex annotation structures (such as segmentation, multi-category, and multi-target). The COCO annotation format (Common Objects in Context) is a widely used image annotation data format, originally proposed by the COCO dataset, and is used to describe data structures in tasks such as object detection, segmentation, and key point detection. It uses JSON files as a storage method, has powerful expression capabilities, and supports the annotation requirements of multiple targets, multiple categories, and complex structures. The COCO format is widely supported, and many deep learning frameworks (such as PyTorch and SSD) provide tools for directly loading COCO data. The standardized data format of the COCO format makes it convenient for developers to manage and exchange annotation data in a unified format. It supports a variety of computer vision tasks such as object detection, instance segmentation, and key point detection. And through detailed and diverse annotations, the model's adaptability to complex scenes is improved. It can also quickly adapt to new tasks or scene requirements by adding new categories or fields.

[0069] In the specific implementation process, in the management of the positioning label frame, the management of the actual hazard source boundary frame is based on absolute coordinates, and a fixed positioning label frame is posted in the hazard source area to assist OpenCV's Canny edge detection algorithm to more accurately and quickly identify and extract the hazard source feature area of ​​the image.

[0070] At the same time, the status of the positioning label frame (such as wear, damage, etc.) needs to be checked and updated regularly. In this way, a clear connection can be established between image recognition and physical object positioning, which not only facilitates the actual management of hazardous sources, but also improves the efficiency and accuracy of OpenCV's recognition.

[0071] Furthermore, it also includes: performing data enhancement (such as rotation, flipping, etc.) on the annotated image to increase the diversity of input samples;

[0072] Furthermore, it also includes: screening verified workshop hazard source images in the OpenCV library and annotating them accordingly as collected images; thereby expanding the sample range of collected images.

[0073] As an optional implementation, the data processing module is used to extract the target region of the acquired image; reduce redundant data to optimize computing resources; and crop the extracted target region (ROI) and convert it into the image input format required by the SSD algorithm.

[0074] The specific process of extracting the target area from the collected image in the data processing module includes:

[0075] Use OpenCV's edge detection algorithm (such as Canny) to parse the bounding box coordinates according to the annotation file and locate the target region (ROI) of the image.

[0076] The target area of ​​the image is further binarized to extract the pixel data of the target area.

[0077] As an optional implementation, it also includes: using OpenCV's super-resolution reconstruction technology to enlarge and enhance the target area, so as to improve the detection model's ability to recognize and analyze the target.

[0078] In the specific implementation process, the data processing module aims to accurately extract the target region of interest (ROI) of the hazard source from the original image, reduce the consumption of computing resources by irrelevant background, and provide efficient and formatted input data for the SSD detection algorithm.

[0079] The data annotation and processing flow of the present invention are all based on the OpenCV software library, which is an open source computer vision and machine learning software library that provides a variety of implementations for image processing, analysis, and machine learning.

[0080] In the specific implementation process, OpenCV provides a graphical interface function that can load images and operate the annotation area through the mouse. When developing customized annotation tools, you can use OpenCV to implement interactive functions. Use OpenCV to read the bounding box coordinates in the annotation file and draw it on the image (such as cv2.rectangle()) for visualization and verification of the accuracy of the annotation.

[0081] Edge detection algorithms in the OpenCV software library (such as Canny) can highlight the boundaries in the image and help users mark the bounding box more accurately. They can also quickly detect the outline of the target and locate the target region (ROI).

[0082] Use OpenCV's 'cv2.imread()' to load and read the raw image.

[0083] Use OpenCV's edge detection (Canny) algorithm to detect significant edge contours in the image by adjusting high and low threshold parameters;

[0084] Use OpenCV's 'cv2.threshold()' to convert the grayscale image into a binary image through threshold segmentation, which helps in the subsequent extraction of pixel features.

[0085] Use OpenCV's 'cv2.crop()' function to crop the target area.

[0086] Therefore, using the efficient image processing functions of the OpenCV software library can significantly accelerate the data annotation process and processing, and optimize the reasoning speed and computational efficiency of subsequent classification algorithms.

[0087] The algorithm training and classification module is used to train the SSD algorithm based on multimodal data to output classification and recognition results;

[0088] Furthermore, the specific process of algorithm training and classification module includes:

[0089] Data preparation: Use COCO format annotation files to load training data;

[0090] SSD algorithm construction: select the basic network of the SSD algorithm (such as VGG-16 or MobileNet); adjust the number of target categories and default box configuration of the output layer according to the scene requirements; use the classification loss function and the positioning loss function as the loss function; use SGD or Adam as the optimizer to optimize the algorithm parameters;

[0091] Write a custom data loader (such as PyTorch’s Dataset class) to read the input image.

[0092] Algorithm verification and optimization: Evaluate the algorithm performance on the validation set and calculate the mean average precision (mean AveragePrecision); adjust hyperparameters (such as learning rate, loss function, optimizer) to optimize the algorithm;

[0093] Algorithm deployment: The newly acquired images are optimized by the data processing module and then input into the trained SSD algorithm, which outputs the image’s hazardous source target category, bounding box coordinates, and environmental data.

[0094] Specifically, the algorithm training and classification module can realize the automatic identification and classification of workshop hazards by training the SSD algorithm; SSD is a single-stage target detection algorithm that can directly predict the target category on the multi-scale feature map at a faster speed. The SSD algorithm is fast, efficient and easy to deploy, and can be closed-loop optimized within a specific small sample range (such as the identification and classification of hazards in various production workshops), thereby greatly improving the efficiency of real-time data processing; meeting the needs of real-time detection and multi-target recognition in industrial production scenarios.

[0095] As a more preferred implementation method, the output results of the algorithm training and classification module (hazard source target category, bounding box coordinates and environmental data) can also be used to supplement the labeled samples of the training data, optimize the labeling of the training data, simplify the data calculation amount, and build a closed-loop computing efficiency optimization system.

[0096] The early warning module is used to generate different levels of warning mechanisms based on the output results and the alarm thresholds of the corresponding target categories.

[0097] Furthermore, the specific process of the early warning module includes:

[0098] Classification management is carried out according to the target category of the image hazard source, specifically:

[0099] When the temperature of the appliance exceeds 60°C, a second-level alarm is triggered; when the temperature of the appliance exceeds 80°C, a first-level alarm is triggered;

[0100] When the shelf structure displacement exceeds 3cm, a second-level alarm is triggered; when the shelf structure displacement exceeds 5cm, a first-level alarm is triggered;

[0101] When a person enters the workshop without wearing safety equipment, a level 2 alarm is triggered; when a person enters the workshop without wearing safety equipment and appears in a high-risk area, a level 1 alarm is triggered;

[0102] When the smoke concentration is detected to be greater than 100ppm, a level 1 alarm is triggered.

[0103] In the specific implementation process, the goal of the early warning module is to generate different levels of warning mechanisms (such as sound and light alarms, SMS notifications) based on the comparison of the hazard source detection results (target category environmental data) with the corresponding alarm thresholds to ensure that potential hazards in the workshop can be discovered and dealt with in a timely manner.

[0104] Alarm level: Level 1 alarm: high risk, requiring emergency treatment (sound and light alarm + SMS notification). Level 2 alarm: medium risk, reminding on-site personnel to pay attention (SMS notification). SMS notification sends the alarm information to security inspectors through the SMS gateway.

[0105] The present invention is based on the production workshop hazard source early warning system of image recognition technology, uses high-definition cameras to collect images, and combines multimodal data such as environmental sensor data (temperature, humidity, smoke concentration); by adopting the coco annotation format and managing image data, the effective fusion of multimodal data is realized; and the image is processed through the OpenCV software library function (edge ​​detection, region extraction, super-resolution reconstruction, etc.), the input data of the SSD algorithm is optimized; the sample training and target detection efficiency of the SSD algorithm are improved; and the algorithm stability and accuracy are improved. Based on the output results of the algorithm, a multi-level early warning mechanism is set to quickly respond to different levels of hazard sources and realize intelligent workshop safety management. At the same time, potential production workshop safety accidents are prevented to ensure the continuity of enterprise production.

[0106] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0107] In the several embodiments provided in the present application, it should be understood that the disclosed modules, processes and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0108] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or software functional modules.

[0109] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

[0110] In the description of the specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0111] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. The production workshop hazard source early warning system based on image recognition technology is characterized by: It includes data collection module, data annotation and management module, detection model training module and early warning module; The data acquisition module is used to deploy high-definition cameras, mobile acquisition devices, and environmental sensors; High-definition cameras are used to collect real-time images of the workshop; The mobile collection device includes an aircraft or a robot vehicle equipped with a camera; the environmental sensor is configured on the mobile collection device and follows the mobile collection device to detect environmental data; The data annotation and management module is used to annotate the collected images using the COCO format, define the basic information, target category, bounding box coordinates, and environmental data of the workshop hazard source images, and store them in a structured directory in the JSON file format to implement classification management. The algorithm training and classification module is used to train the SSD algorithm based on image data to output classification recognition results; The early warning module is used to generate different levels of warning mechanisms based on the output results and the alarm thresholds of the corresponding target categories.

2. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: In the data acquisition module, the environmental sensor includes an infrared thermometer, a smoke sensor or a humidity sensor.

3. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: In the data annotation and management module, the basic information of the workshop hazard source image includes the image file name, image size and unique identifier; Target categories include overheating of electrical appliances, displacement of rack structures, and personnel regulations; The bounding box coordinates are pixel value coordinates; Environmental data include target temperature, humidity, smoke concentration and shelf structure displacement data.

4. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: The data annotation and management module further includes: posting a positioning label frame on the boundary of the specific physical object in the annotated image to implement boundary frame positioning management of the physical object.

5. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: The data labeling and management module also includes: screening verified workshop hazard source images in the OpenCV library and labeling them accordingly as samples of input images.

6. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: A data processing module is also included before the algorithm training and classification module, which is used to extract the target area of ​​the acquired image, crop the extracted target area and convert it into the image input format required by the SSD algorithm.

7. The production workshop hazard source early warning system based on image recognition technology according to claim 6 is characterized by: The specific process of extracting the target area from the collected image in the data processing module includes: Use OpenCV's edge detection algorithm to parse the bounding box coordinates according to the annotation file and locate the target area of ​​the image. The target area of ​​the image is further binarized to extract the pixel data of the target area.

8. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: The specific process of algorithm training and classification module includes: Data preparation: Use COCO format annotation files to load training data; SSD algorithm construction: Select the basic network of the SSD algorithm; adjust the number of target categories and default box configuration of the output layer according to the scene requirements; use the classification loss function and the positioning loss function as the loss function; use SGD or Adam as the optimizer to optimize the algorithm parameters; Write a custom data loader to read the input image; Algorithm verification and optimization: Evaluate algorithm performance on the validation set and calculate the mean average precision; adjust hyperparameters to optimize the algorithm; Algorithm deployment: The newly acquired images are optimized by the data processing module and then input into the trained SSD algorithm, which outputs the image’s hazardous source target category, bounding box coordinates, and environmental data.

9. The production workshop hazard source early warning system based on image recognition technology according to claim 1 is characterized by: The specific process of the early warning module includes: Classification management is carried out according to the target category of the image hazard source, specifically: When the temperature of the appliance exceeds 60°C, a second-level alarm is triggered; when the temperature of the appliance exceeds 80°C, a first-level alarm is triggered; When the shelf structure displacement exceeds 3cm, a second-level alarm is triggered; when the shelf structure displacement exceeds 5cm, a first-level alarm is triggered; When a person enters the workshop without wearing safety equipment, a level 2 alarm is triggered; when a person enters the workshop without wearing safety equipment and appears in a high-risk area, a level 1 alarm is triggered; When the smoke concentration is detected to be greater than 100ppm, a level 1 alarm is triggered.

Images