Mattress spring deformation detection method and system based on image recognition

Abstract

The invention discloses a method and a system for detecting deformation of a spring for a mattress based on image recognition, and relates to the technical field of industrial vision and intelligent detection.According to the method, a multi-camera module, a depth camera, an infrared thermal imaging device and a multi-spectral imaging device are arranged above and on the two sides of a mattress assembly line; collecting mattress surface morphology, indentation distribution, spring geometric depth, temperature and spectral characteristics in real time; calculating a deformation quantity index and carrying out space positioning on the abnormal spring; calculating a thermal spectrum stress anomaly index by combining three-dimensional photometric reconstruction and stress field inversion; the three-dimensional reconstruction spring array obtains a space centroid coordinate and an adjacent distance, and an array consistency coefficient is calculated; and training and detecting log data by using a convolutional neural network, forming an AI dynamic optimization model, and realizing spring health level prediction and abnormal trend analysis. According to the method, the abnormity of the spring can be accurately identified in a multi-dimensional, real-time and accurate manner, local structure optimization and production quality tracing are supported, and the assembly detection precision and efficiency of the mattress are improved.

Description

Technical Field

[0001] This invention relates to the field of industrial vision and intelligent inspection technology, specifically to a method and system for detecting deformation of mattress springs based on image recognition. Background Technology

[0002] As an essential bedding product in modern life, the comfort, support, and durability of a mattress largely depend on the structure and elasticity distribution of its internal spring components. During use, mattress springs may experience compression deformation, eccentricity, tilting, localized collapse, and premature fatigue, all of which directly affect the mattress's support performance and lifespan. However, current mattress production line spring testing methods primarily rely on manual visual inspection or mechanical contact measurement, which have the following technical shortcomings:

[0003] The detection efficiency is low, the degree of automation is insufficient, manual visual inspection is difficult to adapt to the needs of large-scale production, and mechanical contact measurement is complex, time-consuming and inefficient, making it difficult to achieve continuous and real-time monitoring.

[0004] The detection accuracy is limited, making it difficult to detect hidden defects. Manual and traditional mechanical detection methods are easily affected by subjective factors or uneven force application, making it difficult to accurately identify minute compression, tilting or early signs of fatigue of the hidden springs inside the mattress.

[0005] Lacking multimodal information fusion capabilities, the mattress has multiple layers of internal springs, fabrics, fillings, etc. Traditional methods cannot use surface morphology, depth changes, infrared thermal response, and spectral reflectance characteristics for comprehensive analysis, making it difficult to obtain accurate information on the global and local states of the springs.

[0006] The inability to form an intelligent prediction and quality traceability system means that existing testing methods are unable to generate continuous status data, health indices, or trend analysis, and cannot support production process optimization, local structural adjustments, or batch quality traceability. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for detecting the deformation of mattress springs based on image recognition, in order to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a mattress spring deformation detection method based on image recognition, comprising the following steps:

[0009] Step 1: By deploying multiple camera modules, depth cameras, infrared thermal imaging devices, and multispectral imaging devices above and on both sides of the mattress assembly line, the surface morphology, indentation distribution, internal spring geometric depth, temperature distribution, and spectral reflectance characteristics of the mattress are collected in real time. Through spatiotemporal synchronization, brightness equalization, correction, and filtering, a standardized multimodal dataset is formed.

[0010] Step 2: Based on the standardized multimodal dataset, fused depth, infrared and multispectral images, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation exponent SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, identify the abnormal spring space region, apply corresponding strategies and record detection log data.

[0011] Step 3: Perform infrared and multispectral joint re-images of the abnormal spring spatial region, extract stress distribution parameters, temperature difference parameters, and reflectance change parameters through three-dimensional photometric reconstruction and stress field inversion, calculate the thermal spectrum stress anomaly index Hir, and compare it with the thermal spectrum stress anomaly threshold Hth to determine whether the thermal spectrum stress response of the target spring is normal. If it is abnormal, give appropriate strategies and record the detection log data.

[0012] Step 4: Perform 3D reconstruction of the abnormal spring space region, extract the centroid coordinates of the spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, give corresponding strategies and record the detection log data.

[0013] Step 5: Use a convolutional neural network to train the detection log data, extract the feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of the spring, and use it for local structural optimization, production traceability and batch quality assessment.

[0014] Preferably, step one includes:

[0015] S11. A multi-camera module array is set above the mattress assembly line and arranged above and on both sides of the mattress along the direction of the conveyor belt to capture real-time images of the mattress surface fabric and indentation area; RGB image data is collected, including mattress surface morphology, indentation distribution and visible defect information;

[0016] S12. Install depth cameras above and on both sides of the inspection station, and use structured light or binocular stereo imaging to dynamically scan the deformation of the mattress surface and the geometric depth of the hidden layers; collect depth image data, including the height distribution matrix of the mattress surface and the geometric contour projection of the internal spring array.

[0017] S13. Install an infrared thermal imaging device above the mattress conveying path to perform non-contact detection of the temperature distribution in the spring area inside the mattress; collect the micro-heat rise phenomenon generated by the spring under compression or fatigue state, and acquire infrared image data.

[0018] S14. Deploy multispectral imaging devices on both sides of the detection area to perform multi-band scanning of the surface and internal reflection characteristics of the mattress; collect the spectral reflectance changes of the fabric layer and the metal spring layer to obtain multispectral reflectance image data.

[0019] S15. A spatiotemporal synchronous acquisition and control module is used to uniformly align the timestamps of RGB images, depth images, infrared images, and multispectral reflectance images; the acquired data is standardized through brightness equalization algorithm, transmission correction algorithm, and lens distortion correction algorithm; further, adaptive filtering and spatial enhancement technology are used to remove noise interference caused by mattress surface texture and light reflection, forming a standardized multimodal dataset.

[0020] Preferably, step two includes:

[0021] S21. Based on a standardized multimodal dataset, the depth image data, infrared image data, and multispectral reflectance image data of the mattress internal springs are fused and modeled; the three-dimensional geometric parameters of each spring component are extracted through a depth vision reconstruction algorithm, including the detection of spring height hobs and spring central axis tilt angle θ.

[0022] Preferably, step two further includes:

[0023] S22. After dimensionless normalization of the obtained reference spring height href, detected spring height hobs and spring center axis tilt angle θ, the deformation exponent SD is calculated.

[0024] S23. By setting a pre-defined deformation threshold Sth, and comparing the deformation index SD with the deformation threshold Sth, the first evaluation result is obtained, including:

[0025] When the deformation exponent SD < deformation threshold Sth, it indicates that the spring component has a normal shape and no significant compression or tilting deformation. The system maintains the current detection state and continues to monitor.

[0026] When the deformation exponent SD is greater than or equal to the deformation threshold Sth, it indicates that the spring component has an abnormal shape and there is a risk of deformation, triggering the first-level warning instruction and generating the first strategy: perform pixel-level spatial positioning and highlighting of the spring component exceeding the threshold to determine the abnormal spring spatial area; activate the audio-visual prompt device to remind quality inspectors to check the abnormal spring spatial area; start the local re-inspection mechanism to automatically re-acquire infrared and multispectral images of the abnormal spring spatial area to eliminate transient errors; recalculate, and if the deformation exponent SD shows an upward trend, perform local structural re-examination and stress field analysis, and simultaneously record the inspection log data.

[0027] Preferably, step three includes:

[0028] S31. Perform local structural re-image and stress field analysis. Based on the local re-inspection mechanism triggered in step two, perform joint infrared and multispectral detection on the abnormal spring spatial region. Based on the re-acquired infrared and multispectral images of the abnormal spring spatial region, use three-dimensional photometric reconstruction and stress field inversion algorithms to jointly analyze the micro-deformation distribution and temperature gradient field on the target spring surface, and extract the local stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ.

[0029] Preferably, step three further includes:

[0030] S32. After dimensionless normalization of the obtained stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ, the thermal spectral stress anomaly index Hir is calculated.

[0031] S34. By setting a preset thermal spectral stress anomaly threshold Hth, and comparing and analyzing the thermal spectral stress anomaly index Hir with the thermal spectral stress anomaly threshold Hth, the second evaluation results are obtained, including:

[0032] When the thermal spectral stress anomaly index Hir is less than the thermal spectral stress anomaly threshold Hth, it indicates that the thermal spectral stress response of the target spring is normal and should be continuously monitored.

[0033] When the thermal spectral stress anomaly index Hir is greater than or equal to the thermal spectral stress anomaly threshold Hth, it indicates an abnormal thermal spectral stress response of the target spring, posing a risk of early fatigue. This triggers a second warning instruction and generates a second strategy: pixel-level spatial annotation and highlighting of the abnormal target; activation of an audio-visual alert device to remind quality inspectors to review the data; and initiation of a dynamic re-image and thermal-stress joint monitoring mechanism to continuously collect infrared images, multispectral images, and stress field mapping data, updating the thermal spectral stress anomaly index Hir. If the thermal spectral stress anomaly index Hir shows a continuous upward trend during the continuous collection period, the current mattress factory inspection process is suspended, and the entire inspection log data is recorded.

[0034] Preferably, step four includes:

[0035] S41. Based on the depth image data of the standardized multimodal dataset and the abnormal spring spatial region determined in step two, the internal spring array of the mattress is reconstructed in three dimensions to generate a complete three-dimensional point cloud model of the springs; the spatial centroid coordinates of each spring component are extracted, and the spatial distance Δd between the i-th spring and its adjacent springs is calculated using Euclidean distance. i The eccentricity ε of the i-th spring on the horizontal plane is obtained by the least squares fitting plane method. i .

[0036] Preferably, step four further includes:

[0037] S42, the spatial distance Δd between the i-th spring and its adjacent springs is obtained. i The eccentricity ε of the i-th spring on the horizontal plane i After dimensionless normalization, the array consistency coefficient Carr is calculated and obtained.

[0038] S43. By setting a preset array consistency threshold Cth, and comparing the array consistency coefficient Carr with the array consistency threshold Cth, the third evaluation results are obtained, including:

[0039] When the array consistency coefficient Carr ≥ array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is qualified and should be continuously monitored.

[0040] When the array consistency coefficient Carr is less than the array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is unqualified, and there is a risk of overall installation deviation or local stress concentration chain. This triggers the third warning instruction and generates the third strategy: adopting a geometric reconstruction correction algorithm to generate an accurate three-dimensional model through multi-angle depth re-photographing, correcting local eccentricity and spacing errors; and recording spring arrays with a value less than the array consistency threshold Cth in the detection log data.

[0041] Preferably, step five includes:

[0042] S51. Using a convolutional neural network, construct an initial convolutional neural network model, and train and test the initial convolutional neural network model with detection log data. Use the intermediate layer output of the initial model as a feature vector to capture the potential correlation and hidden patterns among the deformation index SD, the thermal spectrum stress anomaly index Hir, and the array consistency coefficient Carr. Use the extracted feature vector to train and test the initial model a second time to form an AI dynamic optimization model. Output the health level and abnormal trend of the mattress springs. Based on the prediction results, optimize the local structure, trace the production process, and evaluate the batch quality.

[0043] Preferably, the image recognition-based mattress spring deformation detection system includes:

[0044] The multimodal data acquisition module is used to deploy multiple camera modules, depth cameras, infrared thermal imaging devices and multispectral imaging devices above and on both sides of the mattress assembly line to collect mattress surface morphology, indentation distribution, internal spring geometric depth, temperature distribution and spectral reflectance characteristics in real time, and form a standardized multimodal dataset through spatiotemporal synchronization, brightness equalization, correction and filtering.

[0045] The deformation assessment module is used to fuse depth, infrared and multispectral images based on a standardized multimodal dataset, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation index SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, it identifies the abnormal spring space region, applies corresponding strategies and records the detection log data.

[0046] The thermal spectral stress analysis module is used to perform infrared and multispectral joint re-images of the abnormal spring spatial region. Through three-dimensional photometric reconstruction and stress field inversion, stress distribution parameters, temperature difference parameters and reflectance change parameters are extracted. The thermal spectral stress anomaly index Hir is calculated and compared with the thermal spectral stress anomaly threshold Hth to determine whether the thermal spectral stress response of the target spring is normal. If it is abnormal, an appropriate strategy is given and the detection log data is recorded.

[0047] The spring array consistency detection module is used to perform three-dimensional reconstruction of abnormal spring space regions, extract the centroid coordinates of spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, an appropriate strategy is given and the detection log data is recorded.

[0048] The AI ​​dynamic optimization model building module is used to train the detection log data using a convolutional neural network, extract feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of springs, and use it for local structural optimization, production traceability, and batch quality assessment.

[0049] This invention provides a method and system for detecting the deformation of mattress springs based on image recognition. It has the following beneficial effects:

[0050] (1) The image recognition-based mattress spring deformation detection method and system, by deploying multiple camera modules, depth cameras, infrared thermal imaging devices and multispectral imaging devices on the mattress assembly line, and by using spatiotemporal synchronization, brightness equalization, correction and filtering to form a standardized multimodal dataset, realizes industrial vision visual measurement and visual inspection, significantly improving the accuracy and reliability of spring deformation recognition.

[0051] (2) The image recognition-based mattress spring deformation detection method and system uses deep vision reconstruction and three-dimensional modeling to extract the spring height and central axis tilt angle, and calculates the deformation index SD and compares it with the threshold to achieve pixel-level spatial positioning and highlighting of abnormal springs, so that the AI ​​vision system can quickly trigger the early warning strategy.

[0052] (3) The image recognition-based mattress spring deformation detection method and system performs infrared and multispectral joint re-images of abnormal spring spatial regions, calculates the thermal spectrum stress anomaly index Hir through three-dimensional photometric reconstruction and modeling kernel analysis, and can detect early fatigue signs and local stress concentration risks in advance. The combination of dynamic re-images and vision processor control strategies improves detection reliability.

[0053] (4) The image recognition-based mattress spring deformation detection method and system uses a convolutional neural network to train the detection log data to form an AI dynamic optimization model, which can simultaneously capture the potential correlation between the deformation index SD, the thermal spectrum stress anomaly index Hir and the array consistency coefficient Carr, predict the spring health level and abnormal trend, realize local structure optimization, production process traceability and batch quality assessment, and improve the overall quality management efficiency of mattress products. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the steps of the image recognition-based mattress spring deformation detection method of the present invention.

[0055] Figure 2 This is a flowchart illustrating the image recognition-based mattress spring deformation detection system of the present invention. Detailed Implementation

[0056] 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.

[0057] Example 1

[0058] Please see Figure 1 This invention provides a method for detecting the deformation of mattress springs based on image recognition, comprising the following steps:

[0059] Step 1: By deploying multiple camera modules, depth cameras, infrared thermal imaging devices, and multispectral imaging devices above and on both sides of the mattress assembly line, the surface morphology, indentation distribution, internal spring geometric depth, temperature distribution, and spectral reflectance characteristics of the mattress are collected in real time. Through spatiotemporal synchronization, brightness equalization, correction, and filtering, a standardized multimodal dataset is formed.

[0060] Step 2: Based on the standardized multimodal dataset, fused depth, infrared and multispectral images, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation exponent SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, identify the abnormal spring space region, apply corresponding strategies and record detection log data.

[0061] Step 3: Perform infrared and multispectral joint re-images of the abnormal spring spatial region, extract stress distribution parameters, temperature difference parameters, and reflectance change parameters through three-dimensional photometric reconstruction and stress field inversion, calculate the thermal spectrum stress anomaly index Hir, and compare it with the thermal spectrum stress anomaly threshold Hth to determine whether the thermal spectrum stress response of the target spring is normal. If it is abnormal, give appropriate strategies and record the detection log data.

[0062] Step 4: Perform 3D reconstruction of the abnormal spring space region, extract the centroid coordinates of the spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, give corresponding strategies and record the detection log data.

[0063] Step 5: Use a convolutional neural network to train the detection log data, extract the feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of the spring, and use it for local structural optimization, production traceability and batch quality assessment.

[0064] In this embodiment, by deploying multiple camera modules, depth cameras, infrared thermal imaging devices, and multispectral imaging devices on the mattress assembly line, a standardized multimodal dataset is formed. Combined with deep vision reconstruction, 3D modeling, and a convolutional neural network dynamic optimization model, real-time detection and prediction of spring deformation, thermal spectral stress, and array consistency are achieved, thereby improving visual measurement accuracy, early anomaly identification capabilities, and production quality traceability efficiency.

[0065] Example 2

[0066] This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, step one includes:

[0067] S11. A multi-camera module array is set above the mattress assembly line and arranged above and on both sides of the mattress along the direction of the conveyor belt to capture real-time images of the mattress surface fabric and indentation area; RGB image data is collected, including mattress surface morphology, indentation distribution and visible defect information;

[0068] S12. Install depth cameras above and on both sides of the inspection station, and use structured light or binocular stereo imaging to dynamically scan the deformation of the mattress surface and the geometric depth of the hidden layers; collect depth image data, including the height distribution matrix of the mattress surface and the geometric contour projection of the internal spring array.

[0069] S13. Install an infrared thermal imaging device above the mattress conveying path to perform non-contact detection of the temperature distribution in the spring area inside the mattress; collect the micro-heat rise phenomenon generated by the spring under compression or fatigue state, and acquire infrared image data.

[0070] S14. Deploy multispectral imaging devices on both sides of the detection area to perform multi-band scanning of the surface and internal reflection characteristics of the mattress; collect the spectral reflectance changes of the fabric layer and the metal spring layer to obtain multispectral reflectance image data.

[0071] S15. A spatiotemporal synchronous acquisition and control module is used to uniformly align the timestamps of RGB images, depth images, infrared images, and multispectral reflectance images; the acquired data is standardized through brightness equalization algorithm, transmission correction algorithm, and lens distortion correction algorithm; further, adaptive filtering and spatial enhancement technology are used to remove noise interference caused by mattress surface texture and light reflection, forming a standardized multimodal dataset.

[0072] In this embodiment, by deploying multiple camera modules, depth cameras, infrared thermal imaging devices, and multispectral imaging devices on the mattress assembly line, and employing spatiotemporal synchronization, brightness equalization, correction, and filtering processes, a standardized multimodal dataset is formed. This enables high-precision, real-time acquisition of mattress surface morphology, indentation distribution, and internal spring geometry and temperature characteristics, providing a reliable data foundation for subsequent deformation analysis and significantly improving the accuracy and efficiency of industrial visual inspection.

[0073] Example 3

[0074] This embodiment is an explanation based on Embodiment 2. Please refer to it. Figure 1 Specifically, step two includes:

[0075] S21. Based on a standardized multimodal dataset, the depth image data, infrared image data, and multispectral reflectance image data of the mattress internal springs are fused and modeled; the three-dimensional geometric parameters of each spring component are extracted through a depth vision reconstruction algorithm, including the detection of spring height hobs and spring central axis tilt angle θ.

[0076] In this embodiment, by fusing depth images, infrared images, and multispectral reflectance images, and using a depth vision reconstruction algorithm to accurately extract the spring height and central axis tilt angle, a high-precision model of the three-dimensional geometry of the springs inside the mattress is achieved, providing accurate data support for subsequent deformation analysis and abnormal spring location.

[0077] Example 4

[0078] This embodiment is an explanation based on Embodiment 3. Please refer to it. Figure 1 Specifically, step two also includes:

[0079] S22. After obtaining the reference spring height href, the detected spring height hobs, and the spring center axis tilt angle θ, and performing dimensionless normalization, the deformation exponent SD is calculated using the following formula:

[0080]

[0081] In the formula, λ represents the attitude weighting coefficient, and href represents the reference spring height;

[0082] The attitude weighting coefficient λ is obtained as follows: Spring samples with different tilt angles are selected in a standard assembly bed. Using a high-precision angle sensor and a three-dimensional deformation testing device, corresponding data sequences of height changes and tilt angles are collected. The influence ratio of angle changes on deformation deviation is obtained using the least squares fitting method. After multiple calibrations and smoothing processes, λ is determined as the coupling sensitivity coefficient between attitude and height deviation, used to characterize the intensity of the influence of attitude tilt on the overall deformation.

[0083] The reference spring height href is obtained by static measurement of an unloaded and uninstalled standard sample spring. Under constant temperature and pressure conditions, its natural height is measured using a laser displacement sensor or a three-dimensional rangefinder. After multi-point sampling and averaging filtering, it is determined as the benchmark value and used as the standard reference height for detecting spring deformation analysis.

[0084] Formula Principle: The deformation exponent SD is used to quantify the degree of geometric deformation of a spring in its assembled state relative to an ideal structure. It comprehensively reflects the compression and attitude deviation of the spring through a weighted sum of the normalized height difference and the attitude tilt term; where the first term... The term λ*|tan(θ)| is used to characterize the high compression ratio of a spring under stress or fatigue conditions, reflecting its longitudinal deformation degree; the second term λ*|tan(θ)| is an attitude correction term, used to compensate for deformation measurement deviations caused by assembly misalignment or tilt angle.

[0085] S23. By setting a pre-defined deformation threshold Sth, and comparing the deformation index SD with the deformation threshold Sth, the first evaluation result is obtained, including:

[0086] When the deformation exponent SD < deformation threshold Sth, it indicates that the spring component has a normal shape and no significant compression or tilting deformation. The system maintains the current detection state and continues to monitor.

[0087] When the deformation exponent SD is greater than or equal to the deformation threshold Sth, it indicates that the spring component has an abnormal shape and there is a risk of deformation, triggering the first-level warning instruction and generating the first strategy: perform pixel-level spatial positioning and highlighting of the spring component exceeding the threshold to determine the abnormal spring spatial area; activate the audio-visual prompt device to remind quality inspectors to check the abnormal spring spatial area; start the local re-inspection mechanism to automatically re-acquire infrared and multispectral images of the abnormal spring spatial area to eliminate transient errors; recalculate, and if the deformation exponent SD shows an upward trend, perform local structural re-examination and stress field analysis, and simultaneously record the inspection log data.

[0088] The deformation threshold Sth is obtained by statistically analyzing the geometric deformation characteristics of standard mattress sample springs under different assembly conditions and stress states. By collecting a large amount of data on spring height changes and central axis tilt angles under static loading, dynamic compression, and long-term fatigue cycle conditions, the distribution range of the deformation index SD is calculated. Combined with material fatigue test data, design tolerances, and quality inspection experience, a reasonable critical judgment value is determined. Finally, a deformation threshold Sth that can distinguish between normal and abnormal deformation states is formed, used for rapid judgment and early warning triggering of the spring's geometric state.

[0089] In this embodiment, by calculating the deformation exponent SD and comparing it with a preset threshold Sth, it is possible to determine in real time whether the shape of the spring component is abnormal, automatically triggering an early warning and local re-inspection mechanism, realizing the accurate positioning and highlighting of abnormal springs, and improving the timeliness and reliability of detecting defects in the internal springs of the mattress.

[0090] Example 5

[0091] This embodiment is an explanation based on Embodiment 4. Please refer to it. Figure 1 Specifically, step three includes:

[0092] S31. Perform local structural re-image and stress field analysis. Based on the local re-inspection mechanism triggered in step two, perform joint infrared and multispectral detection on the abnormal spring spatial region. Based on the re-acquired infrared and multispectral images of the abnormal spring spatial region, use three-dimensional photometric reconstruction and stress field inversion algorithms to jointly analyze the micro-deformation distribution and temperature gradient field on the target spring surface, and extract the local stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ.

[0093] In this embodiment, by performing infrared and multispectral joint re-images of the abnormal spring spatial region, and combining three-dimensional photometric reconstruction and stress field inversion algorithms, the micro-deformation and temperature gradient distribution on the spring surface can be accurately extracted, achieving high-precision analysis of local stress and thermal response, and improving the accuracy of abnormal spring fault identification.

[0094] Example 6

[0095] This embodiment is an explanation based on Embodiment 5. Please refer to it. Figure 1 Specifically, step three also includes:

[0096] S32. After obtaining the stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ, and performing dimensionless normalization, the thermal spectral stress anomaly index Hir is calculated using the following formula:

[0097]

[0098] In the formula, a1, a2 and a3 represent weighting coefficients, Tref represents the reference temperature difference parameter, and σref represents the reference stress distribution parameter;

[0099] a1 = 0.45: Weight of the temperature rise term. Temperature changes directly reflect energy deposition and local stress release, and are usually a sensitive signal of early fatigue. Set as the sovereign weight.

[0100] a2 = 0.35: Weight of the spectral reflectance term. Changes in reflectance are sensitive to surface / interface microcracks and material variations, serving as a minor but important optical compensation factor.

[0101] a3 = 0.2: The weight of the stress term. Stress inversion is greatly affected by model and calibration errors. It serves as an auxiliary but crucial mechanical correction factor.

[0102] The reference temperature difference parameter Tref is obtained by acquiring the steady-state temperature field of the target tissue or test material in thermal equilibrium under conditions without external thermal excitation; using a high-sensitivity infrared thermal imager and thermocouple array, the temperature distribution on the sample surface is simultaneously measured under constant ambient temperature and radiation conditions; after multi-cycle time averaging and spatial filtering, the average value of the stable interval of the temperature distribution is extracted as the reference temperature difference parameter Tref, which is used to characterize the thermal field uniformity and temperature difference reference value under normal conditions.

[0103] The reference stress distribution parameter σref is obtained by calibrating the initial structural state of the target tissue or test sample under no external loading conditions; multi-point stress distribution data are collected under steady-state thermal conditions using a photoelastic stress measuring instrument or strain gauge array; after normalization and statistical smoothing, the average value of the standard deviation of the stress distribution is calculated and determined as the reference stress distribution parameter σref, which is used as a structural stability benchmark for stress change and thermal spectral response analysis.

[0104] Formula principle: First term The first term, RΔ, characterizes the normalized magnitude of the local temperature rise relative to the baseline, reflecting the intensity of the thermal response; the second term, RΔ, characterizes the difference in multispectral reflectance, reflecting the optical response to changes in material surface / interlayer optical properties and microcracks / surface stress; the third term... The normalization level of local stress relative to the benchmark reflects the coupling contribution of mechanical stress concentration or fatigue state to the thermo-spectral response. The weighting coefficients a1, a2 and a3 are used to balance the contributions of the three types of physical quantities in the comprehensive index. They need to be determined through calibration experiments or selected empirically and fine-tuned in implementation. This index linearly integrates thermal, spectral and mechanical information, and can comprehensively determine the local thermo-mechanical anomalies and potential fatigue risks of the spring under multimodal data.

[0105] S34. By setting a preset thermal spectral stress anomaly threshold Hth, and comparing and analyzing the thermal spectral stress anomaly index Hir with the thermal spectral stress anomaly threshold Hth, the second evaluation results are obtained, including:

[0106] When the thermal spectral stress anomaly index Hir is less than the thermal spectral stress anomaly threshold Hth, it indicates that the thermal spectral stress response of the target spring is normal and should be continuously monitored.

[0107] When the thermal spectral stress anomaly index Hir is greater than or equal to the thermal spectral stress anomaly threshold Hth, it indicates an abnormal thermal spectral stress response of the target spring, posing a risk of early fatigue. This triggers a second warning instruction and generates a second strategy: pixel-level spatial annotation and highlighting of the abnormal target; activation of an audio-visual alert device to remind quality inspectors to review the data; and initiation of a dynamic re-image and thermal-stress joint monitoring mechanism to continuously collect infrared images, multispectral images, and stress field mapping data, updating the thermal spectral stress anomaly index Hir. If the thermal spectral stress anomaly index Hir shows a continuous upward trend during the continuous collection period, the current mattress factory inspection process is suspended, and the entire inspection log data is recorded.

[0108] The thermal spectral stress anomaly threshold Hth is obtained by statistically analyzing the thermal-spectral-stress response characteristics of standard mattress springs and their local structures under multimodal excitation conditions. Data on temperature differences, spectral reflectance changes, and stress distribution under no-anomaly, slight fatigue, and expected ultimate load conditions are collected to calculate the distribution range of the thermal spectral stress anomaly index Hir. Combined with material thermodynamic properties, microscopic fatigue behavior, and quality inspection experience, a reasonable critical judgment value is determined. This ultimately forms a threshold Hth that can distinguish between normal and abnormal thermal spectral stress response states, used for intelligent early warning of local spring fatigue or potential damage.

[0109] In this embodiment, by introducing the calculation and threshold determination mechanism of the thermal spectral stress anomaly index Hir, it is possible to achieve highly sensitive identification of early signs of local fatigue in springs based on multimodal detection data. This index comprehensively reflects three types of physical characteristics: temperature change, reflectivity difference, and stress distribution. It can accurately distinguish between normal thermal response and abnormal stress concentration, and can trigger early warning and re-inspection strategies in the early stage of fatigue, thereby effectively avoiding missed detection and misjudgment of structural hidden dangers and improving the intelligence and precision of mattress spring quality inspection.

[0110] Example 7

[0111] This embodiment is an explanation based on Embodiment 6. Please refer to it. Figure 1 Specifically, step four includes:

[0112] S41. Based on the depth image data of the standardized multimodal dataset and the abnormal spring spatial region determined in step two, the internal spring array of the mattress is reconstructed in three dimensions to generate a complete three-dimensional point cloud model of the springs; the spatial centroid coordinates of each spring component are extracted, and the spatial distance Δd between the i-th spring and its adjacent springs is calculated using Euclidean distance. i The eccentricity ε of the i-th spring on the horizontal plane is obtained by the least squares fitting plane method. i .

[0113] In this embodiment, by performing three-dimensional reconstruction of the internal spring array of the mattress and calculating the spatial distance and eccentricity between the springs, the geometric consistency and assembly deviation of each spring in spatial distribution can be accurately quantified. This process can intuitively reflect the overall structural flatness and local installation errors of the spring array, providing a reliable basis for subsequent structural balance analysis and assembly quality assessment, thereby effectively improving the uniformity and stability detection accuracy of the internal support system of the mattress.

[0114] Example 8

[0115] This embodiment is an explanation based on Embodiment 7. Please refer to it. Figure 1 Specifically, step four also includes:

[0116] S42, the spatial distance Δd between the i-th spring and its adjacent springs is obtained. i The eccentricity ε of the i-th spring on the horizontal plane i After dimensionless normalization, the array consistency coefficient Carr is calculated using the following formula:

[0117]

[0118] In the formula, N represents the number of springs being detected, dref represents the reference spring spacing, and γ represents the plane eccentricity correction coefficient;

[0119] The method for obtaining the planar eccentricity correction coefficient γ is as follows: Multiple sets of spring array samples with different eccentricities are arranged on a standard calibration platform. Using a three-dimensional coordinate measuring machine (CMM) and a laser pose sensor, the data on the offset of the center of mass of each spring and the change in the overall spacing error caused by eccentricity are collected. Linear fitting and sensitivity analysis methods are used to calculate the proportion of the influence of eccentricity change on the spacing deviation. After multiple measurements, smoothing and normalization, γ is determined as the correction sensitivity coefficient of planar eccentricity on the array spacing deviation, which is used to characterize the influence intensity of eccentricity on consistency evaluation.

[0120] The reference spring spacing dref is obtained by statically measuring the spatial distance between the center points of adjacent springs in a standard sample bed that is not assembled and subjected to force. Multiple measurements are taken under constant temperature and humidity conditions using a laser rangefinder or high-precision three-dimensional scanning equipment, and the average value is taken. After geometric correction and standardization, it is determined as the reference spring spacing dref, which is used as the benchmark spacing parameter for array consistency calculation.

[0121] Formula Principle: The array consistency coefficient Carr is used to quantify the geometric consistency and assembly balance of the internal spring array of a mattress in terms of spatial distribution. Through a comprehensive calculation of normalized spring pitch deviation and planar eccentricity correction terms, it reflects the uniformity of the overall structure's arrangement and assembly precision; among which, the first term... The primary term, γ*ε, characterizes the ratio of the average deviation of the spring spacing relative to the standard spacing dref, reflecting the structural tightness and regularity of the array in the horizontal direction. When this value is close to 1, it indicates that the springs are evenly distributed and have high structural consistency; conversely, it indicates the existence of local assembly deviations or uneven spacing. i This is a planar eccentricity correction term used to compensate for two-dimensional in-plane structural deviations caused by assembly eccentricity or spring tilt. This term reflects the degree to which the spring's center of mass deviates from the array center on the horizontal plane, thus more accurately assessing the overall geometric consistency of the array. This formula, through the joint correction of spatial geometric errors and planar attitude deviations, achieves a multi-dimensional comprehensive evaluation of the spring array assembly quality, providing a quantitative basis for assembly process optimization and quality control.

[0122] S43. By setting a preset array consistency threshold Cth, and comparing the array consistency coefficient Carr with the array consistency threshold Cth, the third evaluation results are obtained, including:

[0123] When the array consistency coefficient Carr ≥ array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is qualified and should be continuously monitored.

[0124] When the array consistency coefficient Carr is less than the array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is unqualified, and there is a risk of overall installation deviation or local stress concentration chain. This triggers the third warning instruction and generates the third strategy: adopting a geometric reconstruction correction algorithm to generate an accurate three-dimensional model through multi-angle depth re-photographing, correcting local eccentricity and spacing errors; and recording spring arrays with a value less than the array consistency threshold Cth in the detection log data.

[0125] The array consistency threshold Cth is obtained by statistically analyzing the spatial arrangement consistency characteristics of standard mattress spring arrays under different assembly batches and slight eccentricity conditions. By collecting data on the spacing between adjacent springs and horizontal eccentricity from multiple sets of standard sample mattresses, the distribution range of the array consistency coefficient Carr is calculated. Combined with assembly process specifications, geometric tolerances, and quality inspection experience, a reasonable critical judgment value is determined. This ultimately forms the threshold Cth, which can distinguish between acceptable and unacceptable overall array consistency, used for assembly quality assessment and rapid detection of local eccentricity or spacing anomalies.

[0126] In this embodiment, by calculating the array consistency coefficient Carr and comparing it with the preset threshold Cth, the system can automatically quantify and dynamically correct the spatial distribution accuracy of the spring array inside the mattress. When insufficient consistency is detected, the system can actively trigger the geometric reconstruction correction algorithm to compensate for and correct local eccentricity and spacing errors, thereby significantly improving the assembly uniformity and structural stability of the spring array, ensuring the consistency of the overall support performance of the mattress and the deformation reliability of long-term use.

[0127] Example 9

[0128] This embodiment is an explanation based on Embodiment 8. Please refer to it. Figure 1 Specifically, step five includes:

[0129] S51. Using a convolutional neural network, construct an initial convolutional neural network model, and train and test the initial convolutional neural network model with detection log data. Use the intermediate layer output of the initial model as a feature vector to capture the potential correlation and hidden patterns among the deformation index SD, the thermal spectrum stress anomaly index Hir, and the array consistency coefficient Carr. Use the extracted feature vector to train and test the initial model a second time to form an AI dynamic optimization model. Output the health level and abnormal trend of the mattress springs. Based on the prediction results, optimize the local structure, trace the production process, and evaluate the batch quality.

[0130] In this embodiment, by introducing a convolutional neural network modeling mechanism in step five, the intermediate layer feature vector is trained and extracted using detection log data to achieve deep correlation modeling and feature fusion of the deformation index SD, the thermal spectral stress anomaly index Hir, and the array consistency coefficient Carr, thus constructing an AI dynamic optimization model. This model can automatically output the health level and abnormal trend of mattress springs with the support of an image processor or vision processor, significantly improving the accuracy of visual detection and the ability to intelligently analyze data, and realizing the visual evaluation of local structure optimization, production process traceability, and batch quality.

[0131] Example 10

[0132] Please refer to Figure 2 A mattress spring deformation detection system based on image recognition includes:

[0133] The multimodal data acquisition module is used to deploy multiple camera modules, depth cameras, infrared thermal imaging devices and multispectral imaging devices above and on both sides of the mattress assembly line to collect mattress surface morphology, indentation distribution, internal spring geometric depth, temperature distribution and spectral reflectance characteristics in real time, and form a standardized multimodal dataset through spatiotemporal synchronization, brightness equalization, correction and filtering.

[0134] The deformation assessment module is used to fuse depth, infrared and multispectral images based on a standardized multimodal dataset, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation index SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, it identifies the abnormal spring space region, applies corresponding strategies and records the detection log data.

[0135] The thermal spectral stress analysis module is used to perform infrared and multispectral joint re-images of the abnormal spring spatial region. Through three-dimensional photometric reconstruction and stress field inversion, stress distribution parameters, temperature difference parameters and reflectance change parameters are extracted. The thermal spectral stress anomaly index Hir is calculated and compared with the thermal spectral stress anomaly threshold Hth to determine whether the thermal spectral stress response of the target spring is normal. If it is abnormal, an appropriate strategy is given and the detection log data is recorded.

[0136] The spring array consistency detection module is used to perform three-dimensional reconstruction of abnormal spring space regions, extract the centroid coordinates of spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, an appropriate strategy is given and the detection log data is recorded.

[0137] The AI ​​dynamic optimization model building module is used to train the detection log data using a convolutional neural network, extract feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of springs, and use it for local structural optimization, production traceability, and batch quality assessment.

[0138] In this embodiment, by organically combining system modules such as multimodal data acquisition, deformation assessment, thermal-spectral stress analysis, spring array consistency detection, and AI dynamic optimization model, multi-dimensional real-time monitoring and intelligent analysis of mattress spring height, tilt, thermal-spectral stress, and array distribution are achieved. This enables accurate identification of local anomalies, prediction of health trends, and provides a reliable basis for local structural optimization, production traceability, and batch quality assessment, thereby significantly improving the accuracy and efficiency of mattress assembly quality control.

[0139] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0140] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. A method for detecting mattress spring deformation based on image recognition, characterized in that, Includes the following steps: Step 1: By deploying multiple camera modules, depth cameras, infrared thermal imaging devices, and multispectral imaging devices above and on both sides of the mattress assembly line, the surface morphology, indentation distribution, internal spring geometric depth, temperature distribution, and spectral reflectance characteristics of the mattress are collected in real time. Through spatiotemporal synchronization, brightness equalization, correction, and filtering, a standardized multimodal dataset is formed. Step 2: Based on the standardized multimodal dataset, fused depth, infrared and multispectral images, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation exponent SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, identify the abnormal spring space region, apply corresponding strategies and record detection log data. Step 3: Perform infrared and multispectral joint re-images of the abnormal spring spatial region, extract stress distribution parameters, temperature difference parameters, and reflectance change parameters through three-dimensional photometric reconstruction and stress field inversion, calculate the thermal spectrum stress anomaly index Hir, and compare it with the thermal spectrum stress anomaly threshold Hth to determine whether the thermal spectrum stress response of the target spring is normal. If it is abnormal, give appropriate strategies and record the detection log data. Step 4: Perform 3D reconstruction of the abnormal spring space region, extract the centroid coordinates of the spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, give corresponding strategies and record the detection log data. Step 5: Use a convolutional neural network to train the detection log data, extract the feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of the spring, and use it for local structural optimization, production traceability and batch quality assessment.

2. The mattress spring deformation detection method based on image recognition according to claim 1, characterized in that, Step one includes: S11. A multi-camera module array is set above the mattress assembly line and arranged above and on both sides of the mattress along the direction of the conveyor belt to capture real-time images of the mattress surface fabric and indentation area; RGB image data is collected, including mattress surface morphology, indentation distribution and visible defect information; S12. Install depth cameras above and on both sides of the inspection station, and use structured light or binocular stereo imaging to dynamically scan the deformation of the mattress surface and the geometric depth of the hidden layers; collect depth image data, including the height distribution matrix of the mattress surface and the geometric contour projection of the internal spring array. S13. Install an infrared thermal imaging device above the mattress conveying path to perform non-contact detection of the temperature distribution in the spring area inside the mattress; collect the micro-heat rise phenomenon generated by the spring under compression or fatigue state, and acquire infrared image data. S14. Deploy multispectral imaging devices on both sides of the detection area to perform multi-band scanning of the surface and internal reflection characteristics of the mattress; collect the spectral reflectance changes of the fabric layer and the metal spring layer to obtain multispectral reflectance image data. S15. A spatiotemporal synchronous acquisition and control module is used to uniformly align the timestamps of RGB images, depth images, infrared images, and multispectral reflectance images; the acquired data is standardized through brightness equalization algorithm, transmission correction algorithm, and lens distortion correction algorithm; further, adaptive filtering and spatial enhancement technology are used to remove noise interference caused by mattress surface texture and light reflection, forming a standardized multimodal dataset.

3. The mattress spring deformation detection method based on image recognition according to claim 2, characterized in that, Step two includes: S21. Based on a standardized multimodal dataset, the depth image data, infrared image data, and multispectral reflectance image data of the mattress internal springs are fused and modeled; the three-dimensional geometric parameters of each spring component are extracted through a depth vision reconstruction algorithm, including the detection of spring height hobs and spring central axis tilt angle θ.

4. The mattress spring deformation detection method based on image recognition according to claim 3, characterized in that, Step two also includes: S22. After dimensionless normalization of the obtained reference spring height href, detected spring height hobs and spring center axis tilt angle θ, the deformation exponent SD is calculated. S23. By setting a pre-defined deformation threshold Sth, and comparing the deformation index SD with the deformation threshold Sth, the first evaluation result is obtained, including: When the deformation exponent SD < deformation threshold Sth, it indicates that the spring component has a normal shape and no significant compression or tilting deformation. The system maintains the current detection state and continues to monitor. When the deformation exponent SD is greater than or equal to the deformation threshold Sth, it indicates that the spring component has an abnormal shape and there is a risk of deformation, triggering the first-level warning instruction and generating the first strategy: perform pixel-level spatial positioning and highlighting of the spring component exceeding the threshold to determine the abnormal spring spatial area; activate the audio-visual prompt device to remind quality inspectors to check the abnormal spring spatial area; start the local re-inspection mechanism to automatically re-acquire infrared and multispectral images of the abnormal spring spatial area to eliminate transient errors; recalculate, and if the deformation exponent SD shows an upward trend, perform local structural re-examination and stress field analysis, and simultaneously record the inspection log data.

5. The mattress spring deformation detection method based on image recognition according to claim 4, characterized in that, Step three includes: S31. Perform local structural re-image and stress field analysis. Based on the local re-inspection mechanism triggered in step two, perform joint infrared and multispectral detection on the abnormal spring spatial region. Based on the re-acquired infrared and multispectral images of the abnormal spring spatial region, use three-dimensional photometric reconstruction and stress field inversion algorithms to jointly analyze the micro-deformation distribution and temperature gradient field on the target spring surface, and extract the local stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ.

6. The mattress spring deformation detection method based on image recognition according to claim 5, characterized in that, Step three also includes: S32. After dimensionless normalization of the obtained stress distribution parameter σ, temperature difference parameter T, and reflectivity change parameter RΔ, the thermal spectral stress anomaly index Hir is calculated. S34. By setting a preset thermal spectral stress anomaly threshold Hth, and comparing and analyzing the thermal spectral stress anomaly index Hir with the thermal spectral stress anomaly threshold Hth, the second evaluation results are obtained, including: When the thermal spectral stress anomaly index Hir is less than the thermal spectral stress anomaly threshold Hth, it indicates that the thermal spectral stress response of the target spring is normal and should be continuously monitored. When the thermal spectral stress anomaly index Hir is greater than or equal to the thermal spectral stress anomaly threshold Hth, it indicates an abnormal thermal spectral stress response of the target spring, posing a risk of early fatigue. This triggers a second warning instruction and generates a second strategy: pixel-level spatial annotation and highlighting of the abnormal target; activation of an audio-visual alert device to remind quality inspectors to review the data; and initiation of a dynamic re-image and thermal-stress joint monitoring mechanism to continuously collect infrared images, multispectral images, and stress field mapping data, updating the thermal spectral stress anomaly index Hir. If the thermal spectral stress anomaly index Hir shows a continuous upward trend during the continuous collection period, the current mattress factory inspection process is suspended, and the entire inspection log data is recorded.

7. The mattress spring deformation detection method based on image recognition according to claim 6, characterized in that, Step four includes: S41. Based on the depth image data of the standardized multimodal dataset and the abnormal spring spatial region determined in step two, the internal spring array of the mattress is reconstructed in three dimensions to generate a complete three-dimensional point cloud model of the springs; the spatial centroid coordinates of each spring component are extracted, and the spatial distance Δd between the i-th spring and its adjacent springs is calculated using Euclidean distance. i The eccentricity ε of the i-th spring on the horizontal plane is obtained by the least squares fitting plane method. i .

8. The mattress spring deformation detection method based on image recognition according to claim 7, characterized in that, Step four also includes: S42, the spatial distance Δd between the i-th spring and its adjacent springs is obtained. i The eccentricity ε of the i-th spring on the horizontal plane i After dimensionless normalization, the array consistency coefficient Carr is calculated and obtained. S43. By setting a preset array consistency threshold Cth, and comparing the array consistency coefficient Carr with the array consistency threshold Cth, the third evaluation results are obtained, including: When the array consistency coefficient Carr ≥ array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is qualified and should be continuously monitored. When the array consistency coefficient Carr is less than the array consistency threshold Cth, it indicates that the overall installation consistency of the spring array is unqualified, and there is a risk of overall installation deviation or local stress concentration chain. This triggers the third warning instruction and generates the third strategy: adopting a geometric reconstruction correction algorithm to generate an accurate three-dimensional model through multi-angle depth re-photographing, correcting local eccentricity and spacing errors; and recording spring arrays with a value less than the array consistency threshold Cth in the detection log data.

9. The mattress spring deformation detection method based on image recognition according to claim 8, characterized in that, Step five includes: S51. Using a convolutional neural network, construct an initial convolutional neural network model, and train and test the initial convolutional neural network model with detection log data. Use the intermediate layer output of the initial model as a feature vector to capture the potential correlation and hidden patterns among the deformation index SD, the thermal spectrum stress anomaly index Hir, and the array consistency coefficient Carr. Use the extracted feature vector to train and test the initial model a second time to form an AI dynamic optimization model. Output the health level and abnormal trend of the mattress springs. Based on the prediction results, optimize the local structure, trace the production process, and evaluate the batch quality.

10. A mattress spring deformation detection system based on image recognition, applied to the mattress spring deformation detection method based on image recognition as described in any one of claims 1 to 9, characterized in that, include: The multimodal data acquisition module is used to deploy multiple camera modules, depth cameras, infrared thermal imaging devices and multispectral imaging devices above and on both sides of the mattress assembly line to collect mattress surface morphology, indentation distribution, internal spring geometric depth, temperature distribution and spectral reflectance characteristics in real time, and form a standardized multimodal dataset through spatiotemporal synchronization, brightness equalization, correction and filtering. The deformation assessment module is used to fuse depth, infrared and multispectral images based on a standardized multimodal dataset, extract the spring height and central axis tilt angle through depth vision reconstruction, calculate the deformation index SD, and compare it with the deformation threshold Sth to determine whether the spring component shape is normal. If abnormal, it identifies the abnormal spring space region, applies corresponding strategies and records the detection log data. The thermal spectral stress analysis module is used to perform infrared and multispectral joint re-images of the abnormal spring spatial region. Through three-dimensional photometric reconstruction and stress field inversion, stress distribution parameters, temperature difference parameters and reflectance change parameters are extracted. The thermal spectral stress anomaly index Hir is calculated and compared with the thermal spectral stress anomaly threshold Hth to determine whether the thermal spectral stress response of the target spring is normal. If it is abnormal, an appropriate strategy is given and the detection log data is recorded. The spring array consistency detection module is used to perform three-dimensional reconstruction of abnormal spring space regions, extract the centroid coordinates of spring space, obtain the distance between adjacent springs and the horizontal plane eccentricity, calculate the array consistency coefficient Carr, and compare it with the array consistency threshold Cth to determine whether the overall installation consistency of the spring array is qualified. If it is not qualified, an appropriate strategy is given and the detection log data is recorded. The AI ​​dynamic optimization model building module is used to train the detection log data using a convolutional neural network, extract feature vectors of deformation index SD, thermal spectral stress anomaly index Hir, and array consistency coefficient Carr, form an AI dynamic optimization model, predict the health level and abnormal trend of springs, and use it for local structural optimization, production traceability, and batch quality assessment.

Images