A tunnel blasting semi-hole identification method and system based on interference category suppression

Abstract

This invention belongs to the technical field of tunnel blasting half-hole recognition. It proposes a method and system for tunnel blasting half-hole recognition based on interference category suppression. Using tunnel face images as the half-hole recognition object, it does not rely on high-precision 3D point cloud scanning conditions and can adapt to application scenarios where the time window from tunnel blasting to initial support is short and data acquisition is limited. By directly utilizing on-site images for recognition, the threshold for on-site implementation can be reduced, and the applicability of the method in actual construction scenarios can be improved. By labeling interference targets, the half-hole recognition model can learn the differences between half-hole features and non-half-hole features simultaneously during the training phase. In this way, false detections can be reduced under complex backgrounds, weak textures, and local occlusion conditions, thereby improving the reliability of recognition results.

Description

Technical Field

[0001] This invention belongs to the technical field of tunnel blasting half-hole identification, and particularly relates to a method and system for tunnel blasting half-hole identification based on interference category suppression. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The half-hole ratio is often used as an important criterion for measuring the quality of tunnel smooth blasting and the effectiveness of contour control. Currently, when counting the number of half-holes at construction sites, workers mostly rely on manually recording the data after observing the tunnel face. This method is greatly affected by factors such as personal experience, on-site lighting, shooting position, the cleanliness of the tunnel face surface, and local obstruction, resulting in insufficient statistical efficiency and instability of results. It is difficult to meet the requirements of digital management of the construction process for objective, continuous, and traceable indicators.

[0004] From the image characteristics of the tunnel face, the half-holes left after blasting usually appear as narrow, elongated traces or groove-like remnants with a certain extension direction. Their appearance changes significantly with variations in surrounding rock conditions, imaging distance, and lighting conditions, specifically manifested as inconsistent edge sharpness, large variations in visible length, and insufficient local texture continuity. Meanwhile, slender structures such as small conduits, reinforcing bars, and fissures in the tunnel face may also exhibit similar geometric contours and light-dark distributions to the half-holes in the images, easily causing confusion during automatic identification.

[0005] Existing visual recognition research related to blasting focuses more on scenarios such as pre-blast hole positioning, charge assistance, or post-blast block size identification, with relatively limited solutions specifically for automatic half-hole identification at the tunnel face. While existing technologies utilize 3D point cloud information for half-hole identification, which can analyze spatial geometric features, they typically require high data accuracy and stringent scanning conditions. However, the operational window from the completion of tunnel blasting to the initial support is short, and reliable 3D data cannot always be obtained stably on-site. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides a method and system for identifying tunnel blasting half-holes based on interference category suppression. The method unifies the elongated structures that are easily confused with half-holes into the interference category for joint training and identification suppression, thereby improving the accuracy, stability and field applicability of the half-hole identification results.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for identifying tunnel blasting half-holes based on interference category suppression, comprising: Obtain images of the tunnel face after blasting; Rotated bounding boxes are used to label the half-hole targets and interfering targets in the face images to construct a half-hole recognition dataset; the interfering targets are slender structures in the face images that are similar in geometry to the half-hole. The half-hole recognition model is trained using a half-hole recognition dataset. During the training phase, the half-hole category and the interference category are set as detection objects. This allows the half-hole recognition model to learn the features of the half-hole target while learning the features of the slender structure that is similar in appearance but does not belong to the half-hole, thus obtaining the trained half-hole recognition model. The image of the face to be identified is input into the trained half-hole recognition model to obtain the half-hole recognition result.

[0008] Secondly, the present invention provides a tunnel blasting half-hole identification system based on interference category suppression, characterized in that it includes: The acquisition module is configured to acquire images of the tunnel face after blasting. The construction module is configured to: perform bounding box annotation on the half-hole target and interfering targets in the face image to construct a half-hole recognition dataset; the interfering targets are slender structures in the face image that are similar in geometry to the half-hole. The training module is configured to train the half-hole recognition model using the half-hole recognition dataset. During the training phase, the half-hole category and the interference category are set as detection objects, so that the half-hole recognition model learns the features of the half-hole target while learning the features of the slender structure that is similar in appearance but does not belong to the half-hole, thus obtaining the trained half-hole recognition model. The recognition module is configured to input the image of the face to be recognized into the trained half-hole recognition model to obtain the half-hole recognition result.

[0009] Thirdly, the present invention provides an electronic device including a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the method described in the first aspect.

[0010] Fourthly, the present invention provides a computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in the first aspect.

[0011] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0012] The above one or more technical solutions have the following beneficial effects: In this invention, the tunnel face image is used as the object of half-hole identification. It does not rely on high-precision three-dimensional point cloud scanning conditions and can adapt to application scenarios where the time window from tunnel blasting to initial support is short and data acquisition is limited. By directly using on-site images to complete the identification, the threshold for on-site implementation can be reduced and the applicability of the method in actual construction scenarios can be improved. The interference targets are labeled so that the half-hole identification model can learn the differences between half-hole features and non-half-hole features during the training phase. In this way, false detections can be reduced under complex background, weak texture and local occlusion conditions, and the reliability of the identification results can be improved.

[0013] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0014] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0015] Figure 1 This is an overall flowchart of the tunnel blasting half-hole identification method in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the rotating frame annotation and category modeling of the half-hole target and the interfering target in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the half-hole recognition result in Embodiment 1 of the present invention; Among them, 2-1 is a conventionally labeled half-hole target; 2-2 is a half-hole target labeled with a rotated frame; and 2-3 is a label for interfering targets. Detailed Implementation

[0016] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0017] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0018] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0019] Example 1 This embodiment addresses the challenges of identifying half-hole targets at the tunnel face, which are characterized by their elongated shape, significant directional changes, and susceptibility to similar slender interfering structures. To address these characteristics, this embodiment uses post-blast face images as input data, employs a rotating frame to describe the geometric orientation of the half-hole target, and introduces easily confused objects such as small guide pipes, reinforcing bars, and cracks into the interference category. Combined with data augmentation techniques closely mimicking on-site imaging conditions, the model is trained, thereby achieving automatic identification and output of half-hole targets.

[0020] Compared to statistical methods that rely on manual observation, this embodiment improves the consistency and automation of the recognition process. Compared to processing methods that rely on high-precision 3D acquisition, this embodiment can directly utilize images for recognition under conditions of limited on-site time windows, making it more suitable for engineering implementation. Based on this, this embodiment forms a complete technical route consisting of image acquisition, rotated bounding box annotation, category modeling, model training, and on-site recognition.

[0021] Specifically, images of the working face are first acquired after blasting and before initial support to preserve clearer features of the remaining half-holes; then, the half-holes and interfering objects are labeled with rotating bounding boxes to construct a recognition dataset; subsequently, a half-hole recognition model is established based on YOLO12m, and training is completed by combining explicit modeling of interference categories and field condition enhancement strategies; finally, the trained model is used for the recognition and statistical output of newly acquired working face images.

[0022] like Figure 1 As shown, this embodiment provides a method for identifying tunnel blasting half-holes based on interference category suppression, including the following steps: Step S1: Obtain the image of the tunnel face after blasting.

[0023] In step S1, images of the tunnel face are acquired for half-hole identification. After tunnel blasting, ventilation and smoke extraction, muck removal, and initial support are typically carried out sequentially. Since the residual features of the half-hole are easily covered after initial support, it is preferable to complete image acquisition within the time period from the end of muck removal to the start-up mixing process to preserve more complete half-hole contour, shadow, and texture information.

[0024] To improve the adaptability of the subsequent half-aperture recognition model to differences in on-site images, the image acquisition stage does not excessively filter samples based on brightness levels, image sharpness, or framing location. Instead, it retains samples from different shooting distances, angles, lighting conditions, and working face positions. This approach allows subsequent training data to more comprehensively cover the actual on-site imaging conditions.

[0025] In this embodiment, the image samples were taken from the Qingtongshan Tunnel site of the Jinan-Zaozhuang High-Speed ​​Railway, totaling 673 images with resolutions ranging from 667×500 pixels to 4800×2700 pixels. 539 images were used as the training set, and 134 as the validation set. In addition to overall tunnel face images, the samples also included local region images to cover the appearance of the half-hole target at different scales. The training set was mainly used for updating the parameters of the half-hole recognition model, while the validation set was mainly used to verify the recognition performance during training.

[0026] Step S2: Rotate bounding boxes to label the half-hole targets and interfering targets in the face image to construct a half-hole recognition dataset.

[0027] In step S2, such as Figure 2 As shown, the face image obtained in step S1 is labeled to form the half-hole recognition dataset required for subsequent training.

[0028] Labeling is preferably performed on a locally deployed CVAT platform, using a rotating bounding box to label the half-hole target and interfering targets separately. Interfering targets are slender structures in the image that are easily confused with the half-hole, including objects such as small conduits, reinforcing bars, and cracks.

[0029] The reason for using rotated bounding boxes is that half-holes are typically elongated and their orientation is not fixed. If a conventional horizontal bounding box is used, it is easy to include a lot of irrelevant background in the annotation area, which is not conducive to the half-hole recognition model learning the geometric features of the half-hole itself. By using rotated bounding boxes, the center position, scale range, and main extension direction of the target can be recorded more accurately, providing a data foundation for subsequent directional target detection.

[0030] The annotation file can be exported as a structured file for training. In this embodiment, the annotation file is in COCO JSON format and includes at least image correspondence, target category information, and rotated bounding box annotation information, so as to achieve one-to-one matching between image samples and annotation results during training.

[0031] In this embodiment, a rotated bounding box is used instead of a horizontal bounding box to enclose the half-hole target. Since half-holes are usually elongated and have significant directional changes, using a horizontal bounding box often includes a large area of ​​background in the annotation area, causing the half-hole recognition model to learn more information that is not directly related to the half-hole itself. With a rotated bounding box, the long side of the annotation box can better fit the extension direction of the half-hole, thereby improving the accuracy of the target's geometric representation.

[0032] Rotated frame representation not only helps improve positioning accuracy but also helps maintain a stable description of the target contour when the half-aperture pose changes significantly. Even if the same type of half-aperture exhibits different tilt directions at different shooting positions, the rotated frame can still adapt well to its main extension direction, thereby reducing recognition errors caused by unreasonable envelope range.

[0033] Step S3: Establish a half-hole recognition model based on YOLO12m and set the recognition task as a rotating frame detection task.

[0034] In step S3, a half-hole recognition model for the face of the tunnel is established. Considering the requirements for processing speed and recognition accuracy in field applications, this embodiment selects YOLO12m as the basic detection framework. The framework is a one-stage target detection framework, which can complete target localization and category discrimination in a short processing time, making it suitable for the application scenario of this embodiment.

[0035] The image samples formed in step S2, along with their corresponding category labels and rotated bounding box annotations, are input into the detection model. Through training, the half-hole recognition model gradually establishes a mapping relationship between the features of the tunnel face image and the target's position, orientation, and category. After the half-hole recognition model is trained, it can output the position, orientation, category, and confidence score of candidate targets for subsequent filtering, display, and quantity statistics.

[0036] Meanwhile, considering the elongated shape and significant directional changes of the half-hole target, this embodiment sets the recognition task as a rotating bounding box detection task. This allows the half-hole recognition model to learn the main directional features of the target, rather than just learning the horizontal circumscribed region. This approach reduces the influence of the background region on the regression results and improves the accuracy of the half-hole localization results.

[0037] Step S4: Train the half-hole recognition model using a training method that includes explicit modeling of interference categories and enhancement of on-site working conditions.

[0038] In step S4, the half-hole recognition model established in step S3 is trained. During training, the half-hole category and the interference category are both set as detection objects, so that the half-hole recognition model can learn the features of the half-hole target while learning the features of slender structures that are similar in appearance but do not belong to the half-hole category. Through this training arrangement, the model's ability to recognize differences between similar targets can be improved, and the occurrence of false recognition can be reduced.

[0039] The half-holes in the face image are set as a separate target category, while objects with similar slender appearances, such as small pipes, reinforcing bars, and cracks, are grouped into the interference category. Since interference objects may also appear as narrow strips, linear dark lines, or protruding edges in the image, if only half-holes are used as a single recognition category, the model is prone to misclassifying locally similar textures as half-holes.

[0040] In the same facet image, half-hole targets and interference targets may appear separately or simultaneously and be adjacent to each other. When two types of objects are spatially close and have similar appearance features, relying solely on geometric shape for identification can easily lead to misjudgment. This embodiment introduces both half-hole and interference categories during the training phase, enabling the half-hole recognition model to form clearer classification boundaries between similar slender targets.

[0041] During the inference phase, only the half-hole category result is used as the final statistical output, without including interfering categories in the half-hole count. By combining joint learning during the training phase and classification filtering during the inference phase, the reliability of half-hole recognition results in complex backgrounds can be improved.

[0042] In this embodiment, to enhance the adaptability of the half-aperture recognition model to complex on-site images, an on-site condition enhancement strategy is introduced during the training phase. The on-site condition enhancement methods include one or more of the following: random horizontal flipping, random vertical flipping, color dithering, random scaling, random cropping, brightness perturbation, contrast perturbation, motion blur, and local occlusion. Brightness and contrast perturbation are used to simulate image changes under different lighting conditions, motion blur is used to simulate local blurring that may occur during the acquisition process, and local occlusion is used to simulate the impact of dust, equipment components, or local surface coverage on the visibility of the half-aperture. By introducing these changes into the training samples, common lighting fluctuations, local blurring, local occlusion, and viewpoint changes during on-site shooting can be simulated, thereby improving the model's stability in practical applications.

[0043] In this embodiment, the input image size is set to 1024×1024, the batch size is set to 4, the initial learning rate is set to 0.0001, the optimizer is AdamW, and the number of training epochs is set to 150. During training, parameters are updated using training set samples, and the training status and recognition performance are checked using validation set samples to determine whether the trained model meets the requirements for field application.

[0044] In this embodiment, YOLO12m is used as the baseline model, and a half-hole recognition model suitable for half-hole recognition is built on this basis. The focus of improving the half-hole recognition model is not to arbitrarily adjust the network structure out of the application scenario, but to coordinate the design of the target representation method, the interference category modeling method, and the on-site working condition enhancement method in combination with the half-hole recognition task.

[0045] In this embodiment, the half-hole recognition model can use the following total loss function during training: L=λ 1 L box +λ 2 L cls +λ 3 L dfl in, L This represents the total loss during training the half-hole recognition model. L box The rotation box regression loss measures the positional and geometrical deviations between the predicted and actual rotation boxes. L cls The classification loss is used to measure the classification error between the half-hole category and the interference category. L dfl This is the distributed focus loss, used to refine the constraints on the regression results of the target boundary; λ 1. λ 2 and λ 3 represents the weight coefficients of the corresponding loss terms, used to balance the impact of bounding box localization, class discrimination, and boundary refinement on model training. Through the combined effect of the above loss terms, the model can simultaneously consider target localization accuracy and class discrimination ability during training.

[0046] The rotation box regression loss can be expressed as: L box = 1 - IoU r ( B p , B g ) Among them, IoU r ( B p , B g ) for predicting the rotating frame B p With a real rotating frame B g The rotational intersection-union ratio; B p To predict the rotating frame; B g This is a true rotating frame.

[0047] The classification loss can be expressed as:

[0048] in, C Number of categories; y c For the first c The true label of the class; p c For the model to the first c The predicted probability of a class.

[0049] The distribution focus loss can be expressed as:

[0050] in, N The number of parameters involved in the regression; K The number of discrete distribution intervals; q i For the first i The regression parameter at the th... k Target distribution values ​​over discrete intervals; s ik This represents the corresponding probability value predicted by the model.

[0051] Step S5: Input the image of the face to be identified into the trained half-hole recognition model to obtain the half-hole recognition result, and count the number of half-holes when necessary.

[0052] In step S5, such as Figure 3 As shown, newly acquired tunnel face images that were not used in training and validation are input into the trained half-hole recognition model to obtain half-hole recognition results. The recognition results include at least the location, orientation information, and confidence level of the half-hole target, and can generate statistical results of the number of half-holes in the corresponding image when needed.

[0053] The process of generating the recognition results includes: locating and classifying candidate targets in the image to be recognized using a half-aperture recognition model; distinguishing half-aperture targets from interfering targets based on the recognition category; and retaining the half-aperture category results that meet the requirements according to the set confidence level. In this embodiment, the confidence threshold for field application is set to 0.6.

[0054] If the identification results contain both half-hole and interference categories, only the half-hole category is retained as the final statistical object, while the interference category is only used for false detection suppression and does not participate in the quantity statistics. When on-site verification is required, the retained half-hole identification results can be overlaid on the original face image to form a visual identification result map.

[0055] The method in this embodiment is compared with schemes that do not use rotated bounding box representation and schemes that do not introduce interference category modeling. The comparison results show that rotated bounding box representation is beneficial for improving the localization of half-hole targets, while interference category modeling is beneficial for suppressing false detections caused by slender non-target objects such as rebar and cracks; when the two are used in combination, the overall performance of half-hole recognition is further optimized. The above results can be combined with... Figure 3 The recognition results shown are explained.

[0056] In on-site working face image applications, the method of this embodiment can stably identify most half-hole targets and maintain good distinguishing ability against non-target objects such as reinforcing bars and cracks. Therefore, this embodiment can provide a relatively reliable image recognition basis for the statistical analysis of half-hole numbers and the evaluation of smooth blasting quality.

[0057] Example 2 The purpose of this embodiment is to provide a tunnel blasting half-hole identification system based on interference category suppression, including: The acquisition module is configured to acquire images of the tunnel face after blasting. The construction module is configured to: perform bounding box annotation on the half-hole target and interfering targets in the face image to construct a half-hole recognition dataset; the interfering targets are slender structures in the face image that are similar in geometry to the half-hole. The training module is configured to train the half-hole recognition model using the half-hole recognition dataset. During the training phase, the half-hole category and the interference category are set as detection objects, so that the half-hole recognition model learns the features of the half-hole target while learning the features of the slender structure that is similar in appearance but does not belong to the half-hole, thus obtaining the trained half-hole recognition model. The recognition module is configured to input the image of the face to be recognized into the trained half-hole recognition model to obtain the half-hole recognition result.

[0058] In further embodiments, the following is also provided: An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When executed by the processor, the computer instructions perform the method described in Embodiment 1. For brevity, further details are omitted here.

[0059] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0060] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0061] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the method described in Embodiment 1.

[0062] The method in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor. The software modules can reside in readily available storage media in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, a detailed description is not provided here.

[0063] A computer program product includes a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0064] The present invention also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in program modules, which execute in a device on a target real or virtual processor to perform the processes / methods described above. Typically, program modules include routines, programs, libraries, objects, classes, components, data structures, etc., that perform specific tasks or implement specific abstract data types. In various embodiments, the functionality of program modules can be combined or divided among program modules as needed. The machine-executable instructions for the program modules can execute within a local or distributed device. In a distributed device, the program modules can reside in both local and remote storage media.

[0065] The computer program code used to implement the methods of the present invention may be written in one or more programming languages. This computer program code may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the computer or other programmable data processing device, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a computer, partially on a computer, as a stand-alone software package, partially on a computer and partially on a remote computer, or entirely on a remote computer or server.

[0066] In the context of this invention, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, and the like. Examples of signals may include electrical, optical, radio, sound, or other forms of propagation signals, such as carrier waves, infrared signals, etc.

[0067] Those skilled in the art will recognize that the units and algorithm steps described in conjunction with the embodiments herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0068] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for identifying a semi-hole in a tunnel blasting based on interference category suppression, characterized in that, include: Obtain images of the tunnel face after blasting; Rotated bounding boxes were used to annotate the half-hole targets and interfering targets in the face images to construct a half-hole recognition dataset; The interference target is a slender structure in the face image that is similar to the geometry of a semi-aperture; The half-hole recognition model is trained using a half-hole recognition dataset. During the training phase, the half-hole category and the interference category are set as detection objects. This allows the half-hole recognition model to learn the features of the half-hole target while learning the features of the slender structure that is similar in appearance but does not belong to the half-hole, thus obtaining the trained half-hole recognition model. The image of the face to be identified is input into the trained half-hole recognition model to obtain the half-hole recognition result.

2. The tunnel blast half-hole identification method based on interference category suppression according to claim 1, characterized in that, The tunnel face image is acquired within the time window between the muck removal after tunnel blasting and the initial support, and the tunnel face image includes one or both of the following: an overall image of the tunnel face and a local area image.

3. The method for identifying tunnel blasting half-holes based on interference category suppression as described in claim 1, characterized in that, The rotating frame annotation is used to characterize the position, scale, and orientation information of the target, and the long side of the rotating frame is consistent with the main extension direction of the annotated target.

4. The method for identifying tunnel blasting half-holes based on interference category suppression as described in claim 1, characterized in that, It also includes enhancing the face image, and the enhancement methods include one or more of the following: random horizontal flipping, random vertical flipping, color jittering, random scaling, random cropping, brightness perturbation, contrast perturbation, motion blur, and local occlusion.

5. The method for identifying tunnel blasting half-holes based on interference category suppression as described in claim 1, characterized in that, The interference targets include one or more of small conduits, reinforcing bars, and cracks.

6. The method for identifying tunnel blasting half-holes based on interference category suppression as described in claim 1, characterized in that, The acquired images of the tunnel face after blasting include images of the tunnel face after blasting at different shooting distances, from different angles, under different lighting conditions, and at different locations.

7. A tunnel blasting half-hole identification system based on interference category suppression, characterized in that, include: The acquisition module is configured to acquire images of the tunnel face after blasting. The construction module is configured to: perform bounding box annotation on the half-hole target and interfering targets in the face image to construct a half-hole recognition dataset; the interfering targets are slender structures in the face image that are similar in geometry to the half-hole. The training module is configured to train the half-hole recognition model using the half-hole recognition dataset. During the training phase, the half-hole category and the interference category are set as detection objects, so that the half-hole recognition model learns the features of the half-hole target while learning the features of the slender structure that is similar in appearance but does not belong to the half-hole, thus obtaining the trained half-hole recognition model. The recognition module is configured to input the image of the face to be recognized into the trained half-hole recognition model to obtain the half-hole recognition result.

8. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, perform the method described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the method described in any one of claims 1-6.

Images