Tunnel construction monitoring system and monitoring method based on image recognition

Abstract

The invention relates to the technical field of image recognition, and particularly discloses a tunnel construction monitoring system and monitoring method based on image recognition, and the method comprises the following steps: obtaining an image sequence of a monitoring target shot by a monitoring device in a tunnel construction process; identifying an image distortion area from the image sequence; selecting a plurality of feature points and tracking the motion trails of the feature points in the image sequence; distinguishing feature point position change caused by water drop distortion and feature point position change caused by monitoring target deformation based on the motion trail and the position information of the image distortion area; and correcting the position of the feature point based on a distinguishing result, and calculating the deformation of the monitoring target according to the corrected position of the feature point. According to the invention, the accuracy and robustness of construction monitoring in the tunnel are improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and specifically to a tunnel construction monitoring system and method based on image recognition. Background Technology

[0002] Image recognition refers to the technology of using computers to process, analyze, and understand images in order to identify targets and objects of various patterns. It is a practical application of deep learning algorithms.

[0003] In water-rich sections of tunnels or in poorly ventilated environments, the lenses of monitoring devices are prone to condensation of fine water droplets, forming randomly distributed droplets. These adhering droplets refract and scatter incident light, causing local shifts in the imaging light path and resulting in non-uniform optical distortion in the captured images of the surrounding rock or support structure. This distortion manifests as geometric deformations such as stretching and twisting in specific areas. This distortion directly alters the apparent positions of feature points in the image. When algorithms perform displacement analysis and deformation calculations based on distorted images, they may misjudge the coordinate deviations caused by optical distortion as actual structural deformation, leading to systematic distortion of the measurement results. Because the shape, distribution, and refraction effects of water droplets are highly random, and the distortion pattern dynamically changes with water droplet evaporation and condensation, traditional image correction methods based on fixed parameters or global transformations are insufficient to effectively model and eliminate this local time-varying distortion, resulting in a significant reduction in the reliability of deformation monitoring data. Summary of the Invention

[0004] The purpose of this invention is to provide a tunnel construction monitoring system and method based on image recognition, thereby solving the aforementioned technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: A tunnel construction monitoring method based on image recognition includes the following steps: Acquire image sequences of the monitored targets captured by monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence; Multiple feature points are selected outside the distorted region of the image, and the motion trajectory of the feature points in the image sequence is tracked. Based on the motion trajectory and the position information of the image distortion region, the changes in the position of feature points caused by the distortion of water droplets are distinguished from the changes in the position of feature points caused by the deformation of the monitored target. Based on the differentiation results, the position of the feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of the feature points.

[0006] As a further aspect of the present invention: identifying image distortion regions includes: Each image in the image sequence is segmented into multiple sub-regions; Calculate the standard deviation of pixel values ​​in each sub-region. If the standard deviation is less than a preset threshold, the corresponding sub-region is selected as a candidate distortion region. The number of times a sub-region is considered a candidate distortion region within a predetermined time window is counted. If the number of times exceeds a preset limit, the sub-region is considered a distortion region of the image.

[0007] As a further aspect of the present invention: selecting multiple feature points outside the distorted region of the image and tracking the motion trajectory of the feature points in the image sequence, including: The part outside the distorted area of ​​the image is taken as the target part, and feature points are selected in the target part based on the Shi-Tomasi corner detection algorithm; The motion trajectory of feature points is tracked using the Lucas-Kanade optical flow method, and the motion trajectory of each feature point is denoised using a time-domain sliding window averaging filter.

[0008] As a further aspect of the present invention: based on the positional information of the motion trajectory and the distorted region of the image, distinguishing between feature point position changes caused by water droplet distortion and feature point position changes caused by target deformation, including: The denoised motion trajectory is decomposed into low-frequency trend components and high-frequency fluctuation components using a digital filter. The first derivative of the high-frequency fluctuation component is calculated to obtain its rate of change at each time point, and the time interval in which the rate of change continuously exceeds the preset rate of change threshold is identified as a sudden change period. Feature points whose high-frequency fluctuation component amplitude exceeds a preset amplitude threshold during the mutation period are identified as feature points that fluctuate during that period and are recorded as target points. The distance from the target point to its nearest image distortion region is obtained and recorded as the target distance. The maximum value of the high-frequency fluctuation component amplitude of the target point during the abrupt change period is obtained. When the maximum value decreases as the target distance increases, the high-frequency fluctuation component of the target point during the abrupt change period is marked as the change in the position of the feature point caused by the distortion of the water droplet. Apart from the above situations, all other changes in the position of feature points were determined to be caused by deformation of the monitored target.

[0009] As a further aspect of the present invention: correcting the feature point positions based on the differentiation results includes: The high-frequency fluctuation component to be marked is taken as the target component, and its maximum amplitude and the instantaneous fluctuation direction corresponding to the maximum amplitude are extracted from the target component. The positional base correction vector is calculated based on the maximum amplitude and the direction of fluctuation. The magnitude of the base correction vector is proportional to the maximum amplitude, and the direction is opposite to the direction of fluctuation. The base correction vector is multiplied by a scalar multiplication using a distance weighting coefficient to obtain the final correction vector. The distance weighting coefficient is inversely proportional to the target distance. The final correction vector is superimposed with the original coordinates of the target point during the abrupt change period to complete the correction of the feature point position.

[0010] As a further aspect of the present invention: calculating the deformation of the monitored target based on the corrected feature point positions includes: Images acquired during the initial stage of construction are selected as reference benchmarks, and the positions of the feature points that have been corrected in the reference benchmarks are defined as the reference coordinate set. In the image at the current moment, the current coordinates corresponding to each feature point in the reference coordinate set are found by the feature point matching algorithm, thus forming the current coordinate set; Calculate the Euclidean distance for each feature point from the reference coordinate set to the current coordinate set; Statistical analysis was performed on the Euclidean distances of all feature points, and the three-standard-deviation criterion was used to eliminate abnormal Euclidean distances that deviated from the overall distribution pattern. Based on the set of effective Euclidean distances, the average Euclidean distance is calculated as the deformation of the monitored target.

[0011] As a further aspect of the present invention: if the deformation exceeds a preset deformation threshold, a prompt message is sent to a preset administrator.

[0012] A tunnel construction monitoring system based on image recognition, comprising: Acquisition module: Acquires image sequences of the monitored targets captured by the monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence; Localization module: Selects multiple feature points in the area outside the distorted region of the image and tracks the motion trajectory of the feature points in the image sequence; Distinguishing module: Based on the positional information of the motion trajectory and the distorted region of the image, distinguishing between the changes in the position of feature points caused by the distortion of the water droplets and the changes in the position of feature points caused by the deformation of the monitored target; Monitoring module: Based on the differentiation results, the position of feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of feature points.

[0013] The beneficial effects of this invention compared to the prior art are as follows: This invention establishes a spatiotemporally coordinated feature point motion analysis mechanism, effectively suppressing water droplet distortion interference in images under complex humid or water-rich construction environments. By separating the rapid fluctuation component caused by water droplets in the time series and combining it with spatial distribution characteristics for correlation judgment, spurious displacement signals caused by optical distortion can be accurately identified and eliminated. After targeted correction processing, the spatial position of feature points is accurately restored, thus ensuring the stable extraction of true deformation information of the monitored target. This method enables the monitoring system to continuously output highly consistent and reliable deformation data even under conditions of water mist adhesion and light scattering, effectively avoiding misjudgments and measurement deviations caused by local random distortions, improving the reliability of tunnel construction monitoring results and the accuracy of safety early warnings, and possessing the technical advantage of stable application in complex environments such as high water content and low visibility. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a flowchart illustrating a tunnel construction monitoring method based on image recognition according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 As shown, this invention is a tunnel construction monitoring method based on image recognition, comprising the following steps: Acquire image sequences of the monitored targets captured by monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence.

[0018] In a preferred embodiment of the present invention, identifying distorted regions in an image includes: Each image in the image sequence is segmented into multiple sub-regions; Calculate the standard deviation of pixel values ​​in each sub-region. If the standard deviation is less than a preset threshold, the corresponding sub-region is selected as a candidate distortion region. The number of times a sub-region is considered a candidate distortion region within a predetermined time window is counted. If the number of times exceeds a preset limit, the sub-region is considered a distortion region of the image.

[0019] Multiple feature points are selected outside the distorted region of the image, and the motion trajectory of the feature points in the image sequence is tracked. In another preferred embodiment of the present invention, multiple feature points are selected in the portion outside the distorted region of the image, and the motion trajectory of the feature points in the image sequence is tracked, including: The part outside the distorted area of ​​the image is taken as the target part, and feature points are selected in the target part based on the Shi-Tomasi corner detection algorithm; The motion trajectory of feature points is tracked using the Lucas-Kanade optical flow method, and the motion trajectory of each feature point is denoised using a time-domain sliding window averaging filter.

[0020] Based on the motion trajectory and the position information of the image distortion region, the changes in the position of feature points caused by the distortion of water droplets are distinguished from the changes in the position of feature points caused by the deformation of the monitored target. In another preferred embodiment of the present invention, based on the positional information of the motion trajectory and the distorted region of the image, the method distinguishes between feature point position changes caused by water droplet distortion and feature point position changes caused by deformation of the monitored target, including: The denoised motion trajectory is decomposed into low-frequency trend components and high-frequency fluctuation components using a digital filter. The first derivative of the high-frequency fluctuation component is calculated to obtain its rate of change at each time point, and the time interval in which the rate of change continuously exceeds the preset rate of change threshold is identified as a sudden change period. Feature points whose high-frequency fluctuation component amplitude exceeds a preset amplitude threshold during the mutation period are identified as feature points that fluctuate during that period and are recorded as target points. The distance from the target point to its nearest image distortion region is obtained and recorded as the target distance. The maximum value of the high-frequency fluctuation component amplitude of the target point during the abrupt change period is obtained. When the maximum value decreases as the target distance increases, the high-frequency fluctuation component of the target point during the abrupt change period is marked as the change in the position of the feature point caused by the distortion of the water droplet. Apart from the above situations, all other changes in the position of feature points were determined to be caused by deformation of the monitored target.

[0021] It should be noted that the trajectory of the feature point is composed of a superposition of slow structural deformation signals and instantaneous optical disturbance signals, which differ significantly in frequency characteristics. Digital filters separate the spectral components of the time-series signal to distinguish the trends of change within different frequency ranges. During the filtering process, the trajectory data is converted into a continuous-time signal. The filter divides the signal into passband and stopband based on a set cutoff frequency, preserving the slowly changing low-frequency components while extracting the rapidly changing high-frequency components. The low-frequency trend component represents the stable deformation trajectory of the feature point as construction progresses, while the high-frequency fluctuation component reflects short-term random jitter caused by water droplet refraction. The digital filter achieves this separation because it utilizes the differences in the time-domain response of different frequency components in signal processing. By adjusting the weights of the input sequence, it suppresses rapidly changing parts or enhances smoothly changing parts, thereby decomposing the original trajectory into independent signals reflecting different physical sources, providing a basis for subsequent differentiation and correction.

[0022] By performing frequency domain decomposition on the motion trajectory of the denoised feature points, the overall motion signal can be distinguished into low-frequency trend components and high-frequency fluctuation components. This allows for the separation of slow, continuous structural deformation from instantaneous, irregular optical disturbances in terms of temporal characteristics. The first derivative of the high-frequency fluctuation component reflects the rate of change of the feature point's position. When the rate of change continuously exceeds a threshold within a short period, it indicates the presence of a sudden disturbance. If such disturbances occur in the spatial range near the image distortion area, and their fluctuation amplitude decreases with increasing distance from the distortion area, it indicates that the change originates from the refraction or scattering effect of water droplets. Utilizing this principle to determine the trend of high-frequency component changes allows for the identification and differentiation of spurious displacements caused by water droplets from genuine structural deformation in dynamic environments. After this differentiation, the data used for subsequent position correction and deformation calculation retains only the effective displacement information after the removal of water droplet influence. This ensures that even in the presence of local optical distortion, the true deformation state of the monitored target can still be accurately reflected, maintaining the stability and reliability of the measurement results.

[0023] When water droplets adhere to the lens surface, they form a microlens structure with an irregular curved surface, causing local refraction shifts when incident light passes through. As the surface tension of the water droplet changes and the evaporation and condensation processes occur, the angle of light refraction changes abruptly within a very short time, resulting in a rapid and transient jump in the imaging point's position. This jump manifests as high-frequency fluctuations in the time domain, with the rate of change continuously exceeding a threshold, reflecting the sudden disturbance caused by the instantaneous shift in the light path. Furthermore, the optical effect of the water droplet exhibits significant spatial locality; the refractive intensity decreases rapidly with distance from the droplet's center. Therefore, the fluctuation amplitude is larger near the distortion region, and as the distance increases, the light propagation tends to stabilize, and the fluctuation amplitude weakens accordingly. Thus, the continuous abrupt increase in the rate of change reveals the abrupt refraction process caused by the dynamic deformation of the water droplet surface, while the spatial distribution characteristic of the fluctuation amplitude decreasing with distance directly reflects the local effect of optical distortion. Based on this spatiotemporal correlation principle, it can be reliably determined that the sudden high-frequency disturbance originates from the refraction and scattering effect of the water droplet, thereby distinguishing it from the slow, continuous displacement caused by the deformation of the monitored target due to force.

[0024] Based on the differentiation results, the position of the feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of the feature points.

[0025] In another preferred embodiment of the present invention, the correction of feature point positions based on the differentiation results includes: The high-frequency fluctuation component to be marked is taken as the target component, and its maximum amplitude and the instantaneous fluctuation direction corresponding to the maximum amplitude are extracted from the target component. The positional base correction vector is calculated based on the maximum amplitude and the direction of fluctuation. The magnitude of the base correction vector is proportional to the maximum amplitude, and the direction is opposite to the direction of fluctuation. The base correction vector is multiplied by a scalar multiplication using a distance weighting coefficient to obtain the final correction vector. The distance weighting coefficient is inversely proportional to the target distance. The final correction vector is superimposed with the original coordinates of the target point during the abrupt change period to complete the correction of the feature point position.

[0026] It is important to note that water droplets, when adhering to the lens surface, form an irregularly curved microlens structure. When incident light passes through this surface, it undergoes local refraction and shift, causing a momentary displacement of the image point on the image plane. As the surface morphology of the water droplet dynamically changes due to evaporation, condensation, and vibration, the angle of light refraction also changes abruptly within a very short time, resulting in high-frequency fluctuations in the trajectory of the feature point. The instantaneous direction of these fluctuations reflects the displacement direction of the image point on the imaging plane caused by the water droplet refraction disturbance. This direction is independent of the actual deformation direction of the monitored target and only represents a spurious displacement caused by optical distortion. To eliminate the interference of this refraction shift on the feature point coordinates, the direction of the basic correction vector must be opposite to the fluctuation direction. By applying an equal amount of reverse compensation, the shift caused by the optical disturbance is offset, restoring the feature point position to its true coordinates under ideal imaging conditions. Because the influence range of the water droplet refraction effect exhibits significant spatial locality, its interference intensity rapidly decreases with increasing distance from the droplet center. Therefore, a weighting coefficient inversely proportional to the distance is introduced during correction to ensure that the compensation amount conforms to the spatial attenuation law of optical perturbation. This effectively corrects significant deviations near the distortion region while avoiding over-correction of distant regions. By combining this inverse cancellation with distance weighting, the true spatial position of feature points can be recovered even in the presence of water droplet optical distortion, providing a stable and reliable data foundation for subsequent deformation calculations.

[0027] In another preferred embodiment of the present invention, calculating the deformation of the monitored target based on the corrected feature point positions includes: Images acquired during the initial stage of construction are selected as reference benchmarks, and the positions of the feature points that have been corrected in the reference benchmarks are defined as the reference coordinate set. In the image at the current moment, the current coordinates corresponding to each feature point in the reference coordinate set are found by the feature point matching algorithm, thus forming the current coordinate set; Calculate the Euclidean distance for each feature point from the reference coordinate set to the current coordinate set; Statistical analysis was performed on the Euclidean distances of all feature points, and the three-standard-deviation criterion was used to eliminate abnormal Euclidean distances that deviated from the overall distribution pattern. Based on the set of effective Euclidean distances, the average Euclidean distance is calculated as the deformation of the monitored target.

[0028] It is understandable that, after correction, the spatial coordinates of feature points can accurately reflect the morphological state of the monitored target at different times. By using images from the initial construction phase as a reference, the original positions of feature points before deformation can be obtained. The coordinate changes of corresponding feature points in the current image reflect the displacement of the target in physical space; therefore, calculating the Euclidean distance between the coordinates at two different times directly characterizes the local deformation. Euclidean distance, based on linear distance in geometric space, accurately expresses the overall displacement amplitude of each feature point due to force or structural changes. Since individual points may exhibit abnormal shifts due to changes in illumination, occlusion, or matching errors during actual monitoring, statistical analysis and the elimination of outliers using a three-standard-deviation criterion can eliminate the interference of occasional errors on the overall assessment, making the remaining data distribution more representative of the true deformation characteristics. Calculating the average value based on the effective distance set yields a quantitative result of the overall deformation trend, thus reflecting the comprehensive deformation degree of the monitored target in the time dimension in a stable statistical manner, providing a reliable basis for structural safety assessment during tunnel construction.

[0029] If the deformation exceeds a preset deformation threshold, a notification message will be sent to the preset administrator.

[0030] When water droplets adhere to the lens surface, they alter the propagation path of locally incident light, causing random displacement deviations in local pixels during imaging, thus resulting in spurious changes in the coordinates of feature points. To accurately extract deformations of the true structure, the key lies in separating the non-realistic components caused by optical refraction from the mixed signal and compensating for their effects.

[0031] This scheme first identifies concentrated areas of optical distortion by utilizing image texture variation patterns and temporal stability. Spatial division is then used to pinpoint the interference range caused by water droplets, providing a spatial reference for subsequent signal analysis. Subsequently, the motion trajectory of feature points is tracked in areas not directly distorted. The time-series signal is digitally filtered and decomposed into low-frequency trends and high-frequency fluctuations. The low-frequency component reflects the slow deformation of the monitored target, while the high-frequency component reflects the instantaneous jumps caused by light refraction. By analyzing the negative correlation between the amplitude of high-frequency fluctuations and the distance to the distorted area, the source of the disturbance can be determined using the principle of optical energy attenuation. Once optical interference is identified, a reverse correction vector is established based on its fluctuation direction, and distance weighting is applied using spatial attenuation principles to achieve inverse compensation for image point offset, restoring the feature point coordinates to their true positions unaffected by refraction. The deformation of the corrected feature points is then calculated using Euclidean distance. After statistically eliminating anomalies, a stable overall deformation result is obtained. The entire process is based on the geometry of light propagation and refraction. By separating the time-domain frequency characteristics and correlating the spatial position, optical distortion can be identified and corrected, thereby effectively eliminating measurement errors caused by water droplet distortion and ensuring that the monitoring data can truly reflect the structural deformation state.

[0032] A tunnel construction monitoring system based on image recognition, comprising: Acquisition module: Acquires image sequences of the monitored targets captured by the monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence; Localization module: Selects multiple feature points in the area outside the distorted region of the image and tracks the motion trajectory of the feature points in the image sequence; Distinguishing module: Based on the positional information of the motion trajectory and the distorted region of the image, distinguishing between the changes in the position of feature points caused by the distortion of the water droplets and the changes in the position of feature points caused by the deformation of the monitored target; Monitoring module: Based on the differentiation results, the position of feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of feature points.

[0033] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A tunnel construction monitoring method based on image recognition, characterized in that, Includes the following steps: Acquire image sequences of the monitored targets captured by monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence; Multiple feature points are selected outside the distorted region of the image, and the motion trajectory of the feature points in the image sequence is tracked. Based on the motion trajectory and the position information of the image distortion region, the changes in the position of feature points caused by the distortion of water droplets are distinguished from the changes in the position of feature points caused by the deformation of the monitored target. Based on the differentiation results, the position of the feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of the feature points.

2. The tunnel construction monitoring method based on image recognition according to claim 1, characterized in that, Identifying distorted regions in an image includes: Each image in the image sequence is segmented into multiple sub-regions; Calculate the standard deviation of pixel values ​​in each sub-region. If the standard deviation is less than a preset threshold, the corresponding sub-region is selected as a candidate distortion region. The number of times a sub-region is considered a candidate distortion region within a predetermined time window is counted. If the number of times exceeds a preset limit, the sub-region is considered a distortion region of the image.

3. The tunnel construction monitoring method based on image recognition according to claim 1, characterized in that, Multiple feature points are selected outside the distorted region of the image, and the motion trajectory of the feature points in the image sequence is tracked, including: The part outside the distorted area of ​​the image is taken as the target part, and feature points are selected in the target part based on the Shi-Tomasi corner detection algorithm; The motion trajectory of feature points is tracked using the Lucas-Kanade optical flow method, and the motion trajectory of each feature point is denoised using a time-domain sliding window averaging filter.

4. The tunnel construction monitoring method based on image recognition according to claim 1, characterized in that, Based on the positional information of the motion trajectory and the distorted region of the image, the system distinguishes between feature point position changes caused by water droplet distortion and feature point position changes caused by target deformation, including: The denoised motion trajectory is decomposed into low-frequency trend components and high-frequency fluctuation components using a digital filter. The first derivative of the high-frequency fluctuation component is calculated to obtain its rate of change at each time point, and the time interval in which the rate of change continuously exceeds the preset rate of change threshold is identified as a sudden change period. Feature points whose high-frequency fluctuation component amplitude exceeds a preset amplitude threshold during the mutation period are identified as feature points that fluctuate during that period and are recorded as target points. The distance from the target point to its nearest image distortion region is obtained and recorded as the target distance. The maximum value of the high-frequency fluctuation component amplitude of the target point during the abrupt change period is obtained. When the maximum value decreases as the target distance increases, the high-frequency fluctuation component of the target point during the abrupt change period is marked as the change in the position of the feature point caused by the distortion of the water droplet. Apart from the above situations, all other changes in the position of feature points were determined to be caused by deformation of the monitored target.

5. The tunnel construction monitoring method based on image recognition according to claim 4, characterized in that, Correcting the feature point positions based on the differentiation results includes: The high-frequency fluctuation component to be marked is taken as the target component, and its maximum amplitude and the instantaneous fluctuation direction corresponding to the maximum amplitude are extracted from the target component. The positional base correction vector is calculated based on the maximum amplitude and the direction of fluctuation. The magnitude of the base correction vector is proportional to the maximum amplitude, and the direction is opposite to the direction of fluctuation. The base correction vector is multiplied by a scalar multiplication using a distance weighting coefficient to obtain the final correction vector. The distance weighting coefficient is inversely proportional to the target distance. The final correction vector is superimposed with the original coordinates of the target point during the abrupt change period to complete the correction of the feature point position.

6. The tunnel construction monitoring method based on image recognition according to claim 1, characterized in that, The deformation of the monitored target is calculated based on the corrected feature point locations, including: Images acquired during the initial stage of construction are selected as reference benchmarks, and the positions of the feature points that have been corrected in the reference benchmarks are defined as the reference coordinate set. In the image at the current moment, the current coordinates corresponding to each feature point in the reference coordinate set are found by the feature point matching algorithm, thus forming the current coordinate set; Calculate the Euclidean distance for each feature point from the reference coordinate set to the current coordinate set; Statistical analysis was performed on the Euclidean distances of all feature points, and the three-standard-deviation criterion was used to eliminate abnormal Euclidean distances that deviated from the overall distribution pattern. Based on the set of effective Euclidean distances, the average Euclidean distance is calculated as the deformation of the monitored target.

7. The tunnel construction monitoring method based on image recognition according to claim 6, characterized in that, If the deformation exceeds the preset deformation threshold, a prompt message will be sent to the preset administrator.

8. A tunnel construction monitoring system based on image recognition, characterized in that, include: Acquisition module: Acquires image sequences of the monitored targets captured by the monitoring equipment during tunnel construction; Identify areas of image distortion caused by tiny water droplets formed when water mist adheres to the lens from the image sequence; Localization module: Selects multiple feature points in the area outside the distorted region of the image and tracks the motion trajectory of the feature points in the image sequence; Distinguishing module: Based on the positional information of the motion trajectory and the distorted region of the image, distinguishing between the changes in the position of feature points caused by the distortion of the water droplets and the changes in the position of feature points caused by the deformation of the monitored target; Monitoring module: Based on the differentiation results, the position of feature points is corrected, and the deformation of the monitored target is calculated based on the corrected position of feature points.

Images