Method, device, medium and equipment for detecting items containing flammable markings

Abstract

The present invention discloses a method, apparatus, medium, and device for detecting objects containing flammable markings, relating to the field of target detection technology. The method comprises: adding an R-MCAD module between the Up-sampling module and the Concat module of the neck network in an original YOLOv10 network to form an improved neck network, wherein the R-MCAD module includes a BAM attention mechanism module and a residual learning module; constructing an improved YOLOv10 network comprising a backbone network, an improved neck network, and a head network in the original YOLOv10 network; collecting images of objects containing flammable markings to construct a dataset, and using the dataset to train the improved YOLOv10 network to obtain a detection model for detecting objects containing flammable markings. The detection model is then used to detect objects containing flammable markings.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to a method, device, medium and equipment for detecting items containing flammable markings. Background Art

[0002] Against the backdrop of the rapid development of the internet and artificial intelligence (AI), object detection technology has undergone a dramatic shift from manually customized features to deep learning-driven automated processing. As public awareness of safety and environmental protection grows, so too does public expectation for technology to detect items bearing flammable markings. Improper storage and use of items bearing flammable markings can lead to serious safety incidents, such as fires and explosions. Accurate detection technology can help prevent these accidents. To protect these flammable items, many countries and regions have strict regulations regarding the storage, transportation, and use of items bearing flammable markings. Automated detection technology can help businesses and organizations comply with these regulations.

[0003] YOLOv10 is an existing object detection network model that can be used to detect items with flammable markings. However, due to the diversity and size of flammable markings, the generalization and robustness of the original YOLOv10 model for detecting items with flammable markings needs to be improved. Summary of the Invention

[0004] Based on this, in order to solve the technical problems in the prior art, the present invention provides a method, device, medium and equipment for detecting items containing flammable markings.

[0005] The present invention provides a method for detecting an article containing a flammable mark, comprising:

[0006] An R-MCAD module is added between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network, wherein the R-MCAD module includes a BAM attention mechanism module and a residual learning module; an improved YOLOv10 network is constructed including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network;

[0007] Collect images of objects with flammable markings to construct a dataset. Use the dataset to train an improved YOLOv10 network to obtain a detection model for detecting objects with flammable markings.

[0008] The image to be detected is input into the detection model, and the initial feature extraction of the image to be detected is performed through the backbone network to obtain the initial flammable mark feature; the initial flammable mark feature is upsampled through the Up-sampling module in the neck network to obtain the upsampled flammable mark feature; the feature representation of the flammable mark area in the upsampled flammable mark feature is adaptively extracted through the BAM attention mechanism in the R-MCAD module, and keys and values are generated according to the feature representation of the flammable mark area. The feature representation of the flammable mark area is weighted based on the generated keys and values to obtain the weighted flammable mark feature; the upsampled flammable mark feature and the weighted flammable mark feature are fused through the residual learning module in the R-MCAD module to enhance the flammable mark feature transmission and learning in the neck network to obtain the enhanced flammable mark feature; the initial flammable mark feature and the enhanced flammable mark feature are fused through the Concat module to obtain the fused flammable mark feature; the flammable mark object in the fused flammable mark feature is detected through the head network, and the position of the object containing the flammable mark in the image to be processed is output.

[0009] Furthermore, obtaining the weighted flammability identification feature specifically includes:

[0010] Extract the ROI region of interest in the upsampled flammable sign feature map as the feature representation of the flammable sign area;

[0011] The learned weights obtained through training map the ROI region of interest to query, key, and value; where the query is used to represent the feature map related to the flammable identification feature; the key is used to represent all feature maps of the upsampled flammable identification feature; and the value is used to represent the ROI region of interest.

[0012] Use the query to interact with each key to calculate the correlation between them, and normalize all the correlations to obtain the corresponding weighted weights;

[0013] The values are weightedly summed using weighted weights to obtain the output weighted flammable identification feature.

[0014] Furthermore, the enhanced flammable identification feature includes a first enhanced flammable identification feature and a second enhanced flammable identification feature;

[0015] The first enhanced flammable identification feature is output by a first R-MCAD module connected to an output terminal of a first Up-sampling module in the neck network;

[0016] The second enhanced flammable identification feature is output by a second R-MCAD module connected to an output terminal of a second Up-sampling module in the neck network.

[0017] Furthermore, the fused flammable identification feature includes a first fused flammable identification feature and a second fused flammable identification feature;

[0018] The first fused flammable identification feature is output by a first Concat module connected to the output end of the first R-MCAD module in the neck network; the first Concat module is used to splice the second initial flammable identification feature output by the backbone network and the first enhanced flammable identification feature output by the first R-MCAD module;

[0019] The second fused flammable identification feature is output by a second Concat module connected to the output end of the second R-MCAD module in the neck network; the second Concat module is used to splice the first initial flammable identification feature output by the backbone network and the second enhanced flammable identification feature output by the second R-MCAD module.

[0020] The present invention provides a device for detecting items with flammable markings, comprising:

[0021] A model construction module is used to add an R-MCAD module between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network. The R-MCAD module includes a BAM attention mechanism module and a residual learning module. An improved YOLOv10 network is constructed, including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network.

[0022] The model training module is used to collect images of objects with flammable markings to construct a dataset, and use the dataset to train the improved YOLOv10 network to obtain a detection model for detecting objects with flammable markings.

[0023] The detection module inputs the image to be detected into the detection model. The backbone network performs initial feature extraction on the image to be detected, obtaining an initial flammable marker feature. The up-sampling module in the neck network upsamples the initial flammable marker feature to obtain an upsampled flammable marker feature. The BAM attention mechanism in the R-MCAD module adaptively extracts feature representations of the flammable marker region within the upsampled flammable marker feature, generates keys and values based on the feature representations of the flammable marker region, and weights the feature representations of the flammable marker region based on the generated keys and values to obtain a weighted flammable marker feature. The residual learning module in the R-MCAD module fuses the upsampled flammable marker feature with the weighted flammable marker feature to enhance the transfer and learning of flammable marker features in the neck network, obtaining an enhanced flammable marker feature. The concat module fuses the initial flammable marker feature with the enhanced flammable marker feature to obtain a fused flammable marker feature. The head network detects flammable marker objects within the fused flammable marker feature and outputs the location of objects containing flammable markers in the image to be processed.

[0024] The present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements the above-mentioned method for detecting articles containing flammable markings.

[0025] The present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for detecting articles with flammable markings is implemented.

[0026] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects:

[0027] In the method for detecting objects with flammable markings provided by the present invention, an R-MCAD module is added between the Up-sampling module and the Concat module in the neck network of the original YOLOv10. The R-MCAD module is used for weighting before the Concat module is fused. This makes the transferred features more targeted and representative, thereby improving the generalization and robustness of the original YOLOv10 in detecting objects with flammable markings. Specifically:

[0028] The core innovation of the R-MCAD module lies in its use of the BAM (Bottleneck Attention Module) attention mechanism, which can automatically identify and extract key feature information related to flammable signs in a data-dependent manner. The BAM mechanism generates keys and values that are more relevant to flammable signs based on these key feature information. In this way, the degree of attention to different types of flammable signs can be adaptively adjusted, thereby maintaining a high detection accuracy under various sign styles and changing conditions, which improves the generalization performance of the network. The residual learning module in the R-MCAD module uses residual connections to avoid information attenuation or loss during feature transfer, thereby enhancing the feature transfer and learning capabilities, thereby improving the robustness and adaptability of the network when faced with complex changes in flammable sign items. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0030] Figure 1 A schematic flow chart of a method for detecting items with flammable markings provided by the present invention;

[0031] Figure 2 Schematic diagram of the improved YOLOv10 network framework provided by the present invention;

[0032] Figure 3 A schematic diagram of the multi-target test results provided by the present invention;

[0033] Figure 4 This is a schematic diagram of the single-target test results provided by the present invention. DETAILED DESCRIPTION

[0034] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and corresponding drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0035] Accurately detecting items bearing flammable markings is crucial for safety monitoring and risk assessment. Existing object detection algorithms are often affected by image quality and environmental variations when processing flammable markings, resulting in low detection accuracy. Therefore, a more efficient and accurate detection method is needed to improve the recognition rate of flammable objects.

[0036] Based on this, the present invention first uses data preprocessing techniques to standardize and enhance images of flammable items to improve the model's generalization and robustness. Next, an improved YOLOv10 algorithm is used to extract features and locate targets in the images. By introducing a novel attention mechanism and an optimized network structure, YOLOv10 significantly improves the detection performance of flammable signs for small objects and in complex backgrounds. This method is widely applicable to industrial safety monitoring, warehouse management, fire safety, and any other scenarios requiring flammable item detection.

[0037] Example 1

[0038] Figure 1 The flow chart of the method for detecting an item with a flammable mark in this embodiment is shown. Figure 1 The method is described in detail and specifically comprises the following steps:

[0039] S1: Add an R-MCAD module between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network. The R-MCAD module includes a BAM attention mechanism module and a residual learning module. Construct an improved YOLOv10 network including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network.

[0040] The improved YOLOv10 network structure constructed is as follows Figure 2 As shown. An improved YOLOv10 network model is constructed by introducing a new R-MCAD module into the neck network (Neck) of the original YOLOv10. The R-MCAD module contains residual learning, modular channel attention, and diversified strategies to improve the model's detection performance for flammable marked items. Specifically, the R-MCAD module is added between the Up-sampling module and the Concat module in the YOLOv10 neck network. The R-MCAD module contains the BAM (Bottleneck Attention Module) attention mechanism and residual learning. The BAM attention mechanism extracts key information from the data in a data-dependent manner and selects the positions of important keys and values in a data-driven manner. This method is more flexible and enables the module to adaptively adjust the degree of attention to different flammable mark features, thereby improving the model's generalization ability for different types of flammable marks. Residual learning can enhance the transfer and learning process of internal module features, thereby improving the performance and robustness of the entire network for the task of detecting items containing flammable marks.

[0041] Among them, the working steps of the BAM attention mechanism module include: extracting the ROI region of interest in the upsampled flammable logo feature map as the feature representation of the flammable logo area; mapping the ROI region of interest to queries, keys and values through the learnable weights obtained through training; wherein, the query is used to represent the feature map related to the flammable logo feature; the key is used to represent all feature maps in the upsampled flammable logo feature; the value is used to represent the ROI region of interest; using the query to interact with each key to calculate the correlation between them, and normalizing all correlations to obtain the corresponding weighted weights; using the weighted weights to perform weighted summation on the values to obtain the output.

[0042] S2: Collect images of objects with flammable labels to build a dataset, use the dataset to train the improved YOLOv10 network, and obtain a detection model for detecting objects with flammable labels. Specifically, it includes:

[0043] S21: Collect images of objects with flammable markings to build a dataset.

[0044] Image data containing flammable marked items is collected and preprocessed, including grayscale conversion and size standardization. The image data is annotated, including the location and category information of the flammable marked items. Specifically, the sample images are processed using flipping, Gaussian blurring, color adjustment, etc. to obtain expanded images. The collected image data are manually labeled using the Visual Interface Annotation Tool (VIA) so that they conform to the format required by the YOLO target detection algorithm and a corresponding annotated dataset is created. Subsequently, this dataset is divided into two parts: a training set for model training and a test set for evaluating model performance.

[0045] S22: Use the dataset to train the improved YOLOv10 network.

[0046] S221. Resize the image so that it is scaled to a uniform standard size according to the original ratio, and then send the resized image to the neural network for recognition processing.

[0047] S222: The processed image is passed into the backbone network for feature extraction.

[0048] S223. The extracted features are input into the neck network for feature fusion and enhancement again. First, the features extracted from the backbone network are downsampled in the neck network. At the same time, different resolutions are set in the neck network (80x80, 40x40, and 20x20, respectively) to capture features of different scales. These feature layers of different resolutions are then used as inputs to the pyramid attention network structure. The pyramid attention network structure effectively integrates multi-scale feature information through upsampling, R-MCAD module, and channel fusion technology. Finally, the three output branches of the pyramid attention network are sent to the detection head to calculate the loss function.

[0049] S224. After feature enhancement and fusion, the data is fed into a head network with a decoupled structure, in which the classification head and the detection head work independently to optimize the recognition and positioning of the target.

[0050] S225. The model's prediction results are filtered. First, prediction boxes with confidence scores below a certain threshold are removed. Then, the non-maximum suppression (NMS) algorithm is applied to remove overlapping prediction boxes, retaining the prediction boxes most likely to contain the target object. Finally, the coordinates of the prediction boxes are adjusted to the actual size of the image through coordinate transformation to obtain accurate object detection results.

[0051] S226. During the training phase, the quality of the network model can be evaluated through a variety of performance evaluation criteria. When evaluating model performance, we usually refer to the following key indicators: Precision, which measures the proportion of positive categories predicted to be positive; Recall, also known as the recall rate, which measures the proportion of correctly predicted categories among all actual positive categories; Average Precision (AP), which is the average value of the precision at different recall levels, obtained by plotting the Precision-Recall curve and calculating the area under it; and Mean Average Precision (mAP), which is the result of averaging the AP values of all categories. The calculation formulas for each evaluation indicator are as follows:

[0052]

[0053] Where TP represents the number of positive predictions as positive, TN represents the number of positive predictions as negative, FP represents the number of negative predictions as positive, FN represents the number of negative predictions as negative, r(k) represents the kth largest recall, and max(p(r(k))) represents the maximum accuracy at point r. Test results comparing this method with YOLOv10 show significant performance improvements, with improvements of 0.25% higher than YOLOv10 at mAP@0.5% and 4.15% higher than YOLOv10 at mAP@0.5:0.95%.

[0054] S3: Input the image to be detected into the detection model, perform initial feature extraction on the image to be detected through the backbone network to obtain the initial flammable mark feature; upsample the initial flammable mark feature through the Up-sampling module in the neck network to obtain the upsampled flammable mark feature; adaptively extract the feature representation of the flammable mark area in the upsampled flammable mark feature through the BAM attention mechanism in the R-MCAD module, generate keys and values based on the feature representation of the flammable mark area, and weight the feature representation of the flammable mark area based on the generated keys and values to obtain the weighted flammable mark feature; fuse the upsampled flammable mark feature and the weighted flammable mark feature through the residual learning module in the R-MCAD module to enhance the flammable mark feature transmission and learning in the neck network to obtain the enhanced flammable mark feature; fuse the initial flammable mark feature and the enhanced flammable mark feature through the Concat module to obtain the fused flammable mark feature; detect the flammable mark object in the fused flammable mark feature through the head network, and output the position of the object containing the flammable mark in the image to be processed.

[0055] The test set to be predicted is fed into the object detection model, which then outputs the target location. Specifically, the model's predictions are filtered by first removing prediction boxes with confidence scores below a certain threshold. Then, the non-maximum suppression (NMS) algorithm is applied to remove overlapping prediction boxes, retaining the prediction boxes most likely to contain the target object. Ultimately, accurate object detection results are obtained.

[0056] Test results of multiple flammable items such as Figure 3 , the test results of a single flammable marking item are as follows Figure 4 This technology aims to identify flammable materials in real time to enhance their safety. The evaluation results are concise and clear, and can be understood without professional computer knowledge. They are mainly based on two criteria: one is whether the item is flammable; the other is the confidence level, which is the probability that the item belongs to a specific category, and its value is between 0 and 1. Figure 3 and Figure 4 The test results shown are Figure 3 Two targets were identified, both as flammable items with high confidence (over 0.98); Figure 4 An object is identified, which is also a flammable object with high confidence. These results demonstrate the high efficiency of the improved YOLOv10 model in detection.

[0057] based on Figure 1 The method for detecting items with flammable markings shown here improves detection accuracy and speed through algorithm optimization, achieving a lightweight model design that facilitates rapid deployment in resource-constrained environments. Its robustness ensures stable performance under changing environmental conditions, and the automated detection process reduces manual operations and simplifies workflows. This method is easily integrated with existing systems, improving the intelligent level of flammable object monitoring and providing significant cost-effectiveness and safety value. It is suitable not only for professional monitoring scenarios but also for daily safety inspections. Through continuous data training and model iteration, it can continuously improve detection performance, providing a reliable technical means for the safe management of flammable marked items.

[0058] The above is a method for detecting items with flammable markings provided in one or more embodiments of the present invention. Based on the same concept, the present invention also provides a corresponding device for detecting items with flammable markings, including:

[0059] A model construction module is used to add an R-MCAD module between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network. The R-MCAD module includes a BAM attention mechanism module and a residual learning module; and an improved YOLOv10 network is constructed including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network.

[0060] The model training module is used to collect images of objects with flammable labels to construct a dataset, and use the dataset to train the improved YOLOv10 network to obtain a detection model for detecting objects with flammable labels.

[0061] The detection module is used to input the image to be detected into the detection model, perform initial feature extraction on the image to be detected through the backbone network to obtain the initial flammable mark feature; up-sample the initial flammable mark feature through the Up-sampling module in the neck network to obtain the upsampled flammable mark feature; adaptively extract the feature representation of the flammable mark area in the upsampled flammable mark feature through the BAM attention mechanism in the R-MCAD module, and generate keys and values based on the feature representation of the flammable mark area, and weight the feature representation of the flammable mark area based on the generated keys and values to obtain the weighted flammable mark feature; fuse the upsampled flammable mark feature and the weighted flammable mark feature through the residual learning module in the R-MCAD module to enhance the flammable mark feature transmission and learning in the neck network to obtain the enhanced flammable mark feature; fuse the initial flammable mark feature and the enhanced flammable mark feature through the Concat module to obtain the fused flammable mark feature; detect the flammable mark object in the fused flammable mark feature through the head network, and output the position of the object containing the flammable mark in the image to be processed.

[0062] The specific limitations of the device for detecting items with flammable markings can be found in the limitations of the method for detecting items with flammable markings described above and will not be further elaborated here. Each module in the device for detecting items with flammable markings described above can be implemented in whole or in part through software, hardware, or a combination thereof. Each of these modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, allowing the processor to call and execute the corresponding operations of each module.

[0063] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Provided is a method for detecting items containing flammable markings.

[0064] The present invention also provides a computer device structure. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Provided is a method for detecting items containing flammable markings.

[0065] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0066] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present invention.

Claims

1. A method for detecting items containing flammable markings, characterized in that: include: An R-MCAD module is added between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network, wherein the R-MCAD module includes a BAM attention mechanism module and a residual learning module; an improved YOLOv10 network is constructed including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network; Collect images of objects with flammable markings to construct a dataset. Use the dataset to train an improved YOLOv10 network to obtain a detection model for detecting objects with flammable markings. The image to be detected is input into the detection model, and the initial feature extraction of the image to be detected is performed through the backbone network to obtain the initial flammable mark feature; the initial flammable mark feature is upsampled through the Up-sampling module in the neck network to obtain the upsampled flammable mark feature; the feature representation of the flammable mark area in the upsampled flammable mark feature is adaptively extracted through the BAM attention mechanism in the R-MCAD module, and keys and values are generated according to the feature representation of the flammable mark area. The feature representation of the flammable mark area is weighted based on the generated keys and values to obtain the weighted flammable mark feature; the upsampled flammable mark feature and the weighted flammable mark feature are fused through the residual learning module in the R-MCAD module to enhance the flammable mark feature transmission and learning in the neck network to obtain the enhanced flammable mark feature; the initial flammable mark feature and the enhanced flammable mark feature are fused through the Concat module to obtain the fused flammable mark feature; the flammable mark object in the fused flammable mark feature is detected through the head network, and the position of the object containing the flammable mark in the image to be processed is output.

2. The method for detecting articles containing flammable markings according to claim 1, wherein: The obtaining of the weighted flammability identification feature specifically includes: Extract the ROI region of interest in the upsampled flammable sign feature map as the feature representation of the flammable sign area; The learned weights obtained through training map the ROI region of interest to query, key, and value; where the query is used to represent the feature map related to the flammable identification feature; the key is used to represent all feature maps of the upsampled flammable identification feature; and the value is used to represent the ROI region of interest. Use the query to interact with each key to calculate the correlation between them, and normalize all the correlations to obtain the corresponding weighted weights; The values are weightedly summed using weighted weights to obtain the output weighted flammable identification feature.

3. The method for detecting articles containing flammable markings according to claim 1, wherein: The enhanced flammable identification feature includes a first enhanced flammable identification feature and a second enhanced flammable identification feature; The first enhanced flammable identification feature is output by a first R-MCAD module connected to an output terminal of a first Up-sampling module in the neck network; The second enhanced flammable identification feature is output by a second R-MCAD module connected to an output terminal of a second Up-sampling module in the neck network.

4. The method for detecting articles containing flammable markings according to claim 3, wherein: The fused flammable identification feature includes a first fused flammable identification feature and a second fused flammable identification feature; The first fused flammable identification feature is output by a first Concat module connected to the output end of the first R-MCAD module in the neck network; the first Concat module is used to splice the second initial flammable identification feature output by the backbone network and the first enhanced flammable identification feature output by the first R-MCAD module; The second fused flammable identification feature is output by a second Concat module connected to the output end of the second R-MCAD module in the neck network; the second Concat module is used to splice the first initial flammable identification feature output by the backbone network and the second enhanced flammable identification feature output by the second R-MCAD module.

5. A device for detecting items with flammable markings, characterized in that: include: A model construction module is used to add an R-MCAD module between the Up-sampling module and the Concat module of the neck network in the original YOLOv10 network to form an improved neck network. The R-MCAD module includes a BAM attention mechanism module and a residual learning module. An improved YOLOv10 network is constructed, including the backbone network in the original YOLOv10 network, the improved neck network, and the head network in the original YOLOv10 network. The model training module is used to collect images of objects with flammable markings to construct a dataset, and use the dataset to train the improved YOLOv10 network to obtain a detection model for detecting objects with flammable markings. The detection module is used to input the image to be detected into the detection model, perform initial feature extraction on the image to be detected through the backbone network to obtain the initial flammable mark feature; up-sample the initial flammable mark feature through the Up-sampling module in the neck network to obtain the upsampled flammable mark feature; adaptively extract the feature representation of the flammable mark area in the upsampled flammable mark feature through the BAM attention mechanism in the R-MCAD module, and generate keys and values based on the feature representation of the flammable mark area, and weight the feature representation of the flammable mark area based on the generated keys and values to obtain the weighted flammable mark feature; fuse the upsampled flammable mark feature and the weighted flammable mark feature through the residual learning module in the R-MCAD module to enhance the flammable mark feature transmission and learning in the neck network to obtain the enhanced flammable mark feature; fuse the initial flammable mark feature and the enhanced flammable mark feature through the Concat module to obtain the fused flammable mark feature; detect the flammable mark object in the fused flammable mark feature through the head network, and output the position of the object containing the flammable mark in the image to be processed.

6. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 4 is implemented.

Images