System for detecting heat treatment marks on wood tray based on visual detection

Abstract

The invention relates to the technical field of computer vision and image processing, in particular to a system for detecting a heat treatment mark on a wood tray based on visual detection, which comprises an image enhancement unit for performing wood grain suppression and contrast enhancement processing on a region of interest in an acquired wood tray image to obtain an enhanced image; the degradation evaluation unit is used for performing frequency domain energy distribution analysis on the enhanced image to obtain a surface degradation index; and the grading detection unit is used for carrying out grading detection processing on the enhanced image based on the surface degradation index to obtain a mark identification result, and the method dynamically selects and executes a detection process which is most adaptive to the current condition from a plurality of algorithm strategy libraries, and outputs the mark identification result with high confidence.

Description

Technical Field

[0001] This invention relates to the field of computer vision and image processing technology, specifically to a visual detection system for detecting heat treatment marks on wooden pallets. Background Technology

[0002] According to the International Plant Protection Convention (ISPM) 15 guidelines, wood packaging materials for cross-border logistics must be marked with heat treatment. In actual circulation, these marks will degrade to varying degrees due to environmental erosion and physical wear, such as fading and blurring. Existing visual inspection methods mostly use fixed parameter models, which can achieve an accuracy of up to 95% when processing brand new marks. However, when the marks degrade, their fixed image processing strategies cannot adapt to changing image features, resulting in a severe decrease in the distinction between key information and background, and the detection rate drops sharply to below 40%. The core defect is the lack of a quantitative perception mechanism for the current state of the mark, thus making it impossible to dynamically adjust subsequent processing and recognition strategies, resulting in poor detection robustness. Summary of the Invention

[0003] The purpose of this invention is to provide a vision-based detection system for heat treatment marking on wooden pallets, which solves the problems existing in the background art.

[0004] To address the aforementioned technical problems, this invention provides a vision-based detection system for heat treatment markings on wooden pallets, comprising: The image enhancement unit is used to perform wood grain suppression and contrast enhancement processing on the region of interest within the acquired wooden pallet image to obtain an enhanced image; A degradation assessment unit is used to perform frequency domain energy distribution analysis on the enhanced image to obtain a surface degradation index; The hierarchical detection unit is used to perform hierarchical detection processing on the enhanced image based on the surface degradation index to obtain the marker recognition result.

[0005] Preferably, the wood grain suppression and contrast enhancement process is as follows: Calculate the gradient information of each pixel in the region of interest; calculate the wood grain directionality suppression weight based on the gradient information and the degree of deviation from the preset orthogonal direction; obtain the local contrast information of the region of interest; and generate an enhanced image by combining the wood grain directionality suppression weight and the local contrast information.

[0006] Preferably, the frequency domain energy distribution analysis process is as follows: A two-dimensional discrete Fourier transform is performed on the enhanced image to obtain the centered spectrum; the energy of the predefined high-frequency region in the centered spectrum and the total energy of the full spectrum region are calculated; based on the energy of the high-frequency region and the total energy, the surface degradation index is calculated.

[0007] Preferably, the predefined high-frequency region is set according to the higher frequency range outside the characteristic frequency band of the corresponding mark stroke in the feature spectrum of the standard heat-processed mark image.

[0008] Preferably, the graded detection process is as follows: The surface degradation index is compared and analyzed with preset low degradation thresholds and preset high degradation thresholds. When the surface degradation index is less than the preset low degradation threshold, a clear marking signal is generated. When the surface degradation index is greater than or equal to the preset low degradation threshold and less than the preset high degradation threshold, a moderate degradation signal is generated. When the surface degradation index is greater than or equal to the preset high degradation threshold, a severe degradation signal is generated.

[0009] Preferably, when the clear marker signal is generated, the hierarchical detection unit processes the enhanced image using a standard template matching algorithm to obtain the marker recognition result.

[0010] Preferably, when the moderate degradation signal is generated, the hierarchical detection unit uses a lightweight convolutional neural network model to process the enhanced image to obtain the marker recognition result.

[0011] Preferably, when the severe degradation signal is generated, the processing procedure of the grading detection unit is as follows: Generative adversarial networks are used to repair the enhanced image, resulting in a repaired image; a lightweight convolutional neural network model is then used to process the repaired image to obtain the label recognition result.

[0012] Compared with the prior art, the present invention has the following beneficial effects: (1) The image enhancement unit accurately distinguishes the marking strokes from the wood grain background in the preprocessing stage through an innovative wood grain directionality suppression weight function, which greatly suppresses the wood grain as the main source of interference. The degradation evaluation unit introduces a surface degradation index based on the spectral energy distribution, which transforms the image into an objective value that changes continuously within the interval and can be accurately measured. The derivation logic of this index is rooted in Fourier analysis theory, that is, the high-frequency energy of a clear image is much greater than that of a blurry image, so the proportion of high-frequency energy can be used as a direct representation of the degree of degradation. This quantitative evaluation capability is completely absent in the existing technology. It enables the system to accurately perceive the current state of the marking, thereby providing a solid data foundation for subsequent adaptive processing and ensuring that the system can maintain stable and high-precision recognition performance even when the marking is severely degraded.

[0013] (2) Based on the design of its hierarchical detection unit, a complete adaptive processing closed loop is constructed. Based on the comparative analysis results of the aforementioned surface degradation index, the unit can autonomously make decisions and schedule the optimal detection strategy. The system can initiate a deep processing flow that first uses generative adversarial networks for image restoration and then uses convolutional neural networks for recognition. This data-driven dynamic decision-making and hierarchical processing mechanism enables the system to automatically adapt to the real-time state of the marker without any manual intervention, thereby improving the automation level of the detection system and the universality of application scenarios.

[0014] (3) It achieves efficient utilization of computing resources; the design of the hierarchical detection unit ensures that the allocation of computing resources matches the actual difficulty of the task; high computational cost algorithms, such as lightweight convolutional neural network models and generative adversarial networks, are only activated when the system determines that it is necessary to call them when the degree of label degradation is indeed necessary; in the case of a large number of well-labeled images, the system runs a more lightweight and efficient standard template matching algorithm; this refined resource allocation strategy avoids the waste of computing power caused by using a single complex model regardless of the image condition in traditional schemes, and finally achieves the optimization of detection efficiency and hardware cost while ensuring the maximum overall detection accuracy. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a logic block diagram of the system of the present invention; Figure 2 This is a logic block diagram of the wood grain suppression and contrast enhancement processing of the present invention; Figure 3 This is a logic block diagram of the frequency domain energy distribution analysis of this invention; Figure 4 This is a logic block diagram of the predefined high-frequency region of the present invention. Detailed Implementation

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

[0017] Example 1 Please see Figure 1 This invention provides a vision-based detection system for detecting heat treatment marks on wooden pallets, comprising: an image enhancement unit for performing wood grain suppression and contrast enhancement processing on the region of interest within an acquired wooden pallet image to obtain an enhanced image; a degradation evaluation unit for performing frequency domain energy distribution analysis on the enhanced image to obtain a surface degradation index; and a hierarchical detection unit for performing hierarchical detection processing on the enhanced image based on the surface degradation index to obtain a mark recognition result. In this specific embodiment, the vision-based detection system for heat treatment marks on wooden pallets constructs an adaptive detection process that includes perception enhancement, quantitative evaluation, and hierarchical decision-making. The image enhancement unit undertakes the front-end perception enhancement task, its core responsibility being to process images of wooden pallets with complex wood grain backgrounds captured by an industrial camera and output an enhanced image that highlights the mark information while suppressing interference. Subsequently, the degradation evaluation unit performs quantitative analysis on the enhanced image to accurately assess the degree of degradation of the current heat treatment marks due to wear and erosion, generating an objective surface degradation index. Finally, the hierarchical detection unit dynamically selects and executes the detection process best suited to the current situation from a library of various algorithm strategies based on the mark status reflected by this index, outputting a high-confidence mark recognition result. This technical solution aims to solve the technical problem of existing technologies using fixed-parameter models that cannot adapt to mark degradation, leading to a significant decrease in detection rate under complex working conditions.

[0018] Example 2 Please see Figure 2 The wood grain suppression and contrast enhancement process is as follows: calculate the gradient information of each pixel in the region of interest; calculate the wood grain directional suppression weight based on the gradient information and the degree of deviation of the preset orthogonal direction; obtain the local contrast information of the region of interest; and generate an enhanced image by combining the wood grain directional suppression weight and the local contrast information. Please see Figure 3 The frequency domain energy distribution analysis process is as follows: perform a two-dimensional discrete Fourier transform on the enhanced image to obtain the centered spectrum; calculate the energy of the predefined high-frequency region in the centered spectrum and the total energy of the full spectrum region; and calculate the surface degradation index based on the energy of the high-frequency region and the total energy. In this specific embodiment, the image enhancement unit performs refined preprocessing of the original image through a model that couples texture directionality analysis and local illumination compensation. The technical concept lies in the fact that the heat-treated markings and the wood background have fundamentally different texture directions. To utilize this prior knowledge, the processing procedure calculates wood grain directionality suppression weights. This achieves background suppression; the formula for calculating the weight combines the ideas of Gaussian filtering and directional filtering, and its mathematical expression is:

[0019] In the formula, Indicates the gradient magnitude of a pixel; Indicates the gradient direction; It is the expected gradient magnitude of the target stroke obtained through statistical analysis of clearly marked samples; It is an adjustment parameter that controls the sensitivity of gradient magnitude; and These represent the standard horizontal and vertical directions, respectively. Some are used to filter pixels whose gradient strength is close to the ideal marker, while Some pixels are used to suppress pixels whose gradient direction deviates from the standard orthogonal direction, thereby effectively distinguishing the marker from the wood grain. Subsequently, the wood grain directionality suppression weights are fused with local contrast information to generate an enhanced image. This process is an extension of the local adaptive thresholding idea, transforming it from a binary criterion into a grayscale enhancement model, the formula of which is:

[0020] In the formula, These are the original pixel values; It is the average gray level of the local neighborhood; The term represents contrast information after removing the effects of local brightness; it is dimensionless. Weights are applied here to achieve directional enhancement of the marker orientation contrast, and As a global average grayscale addition, it aims to maintain the overall brightness of the enhanced image; after processing by this unit, the output enhanced image is... It features a significantly improved signal-to-noise ratio; The degradation assessment unit receives this The image is analyzed, and a frequency domain energy distribution analysis is initiated. The technical motivation for this analysis stems from the Fourier analysis principle, which states that image sharpness is reflected in the energy of its high-frequency components. To apply this spectral analysis principle to the quantitative assessment of degradation levels, the degradation assessment unit... Perform a two-dimensional discrete Fourier transform and center the spectrum, then base it on the surface degradation index. Calculate using the formula:

[0021] In the formula, For the centered spectrum, the denominator is the entire spectrum region. Internal spectral energy The summation represents the total energy of the image, while the molecules are located in a predefined high-frequency region. Energy accumulation within; The range was normalized to Its value is positively correlated with the degree of degradation; its calculation results provide direct and quantitative decision input for subsequent graded detection and processing.

[0022] Example 3 Please see Figure 4 The predefined high-frequency region is set based on the higher frequency range outside the characteristic frequency band of the corresponding mark stroke in the characteristic spectrum of the standard heat-processed mark image; In this specific embodiment, a predefined high-frequency region is defined. The setting of this region is a crucial technical setting, as it is essential to the accuracy of the surface degradation index. Instead of using a generic high-pass filter, this setting is based on spectral analysis of a large number of standard heat-treated marking samples. Specifically, the spectral energy of a clear marking image is mainly concentrated in a specific mid-frequency band corresponding to characteristics such as the average stroke width and spacing of the marking characters. Therefore, a predefined high-frequency region is crucial. This is defined as a higher frequency region outside the characteristic frequency band; for example, in a centralized spectral coordinate system. It can be defined as satisfying All frequency points The set of cutoff frequencies The upper limit of the characteristic frequency is obtained by performing spectral statistical analysis on a large number of clear marker samples. This setting method based on prior knowledge ensures that the calculation of the surface degradation index can focus on the loss of high-frequency information caused by blurring and breakage of the marker itself, thereby improving the pertinence and robustness of degradation assessment.

[0023] Example 4 The graded detection process is as follows: the surface degradation index is compared and analyzed with preset low degradation threshold and preset high degradation threshold; when the surface degradation index is less than the preset low degradation threshold, a clear marking signal is generated; when the surface degradation index is greater than or equal to the preset low degradation threshold and less than the preset high degradation threshold, a moderate degradation signal is generated; when the surface degradation index is greater than or equal to the preset high degradation threshold, a severe degradation signal is generated. In this specific embodiment, the hierarchical detection unit is the core technical component for realizing the system's adaptive decision-making and execution; it receives the surface degradation index calculated by the degradation assessment unit. and compared it with two preset thresholds—a low degradation threshold. and high degradation threshold —Comparative analysis is performed; its decision-making logic follows preset rules: if Generate a clear marker signal; if Generates a moderately degraded signal; if If this occurs, a severe degradation signal will be generated; to ensure the objectivity and reproducibility of this decision-making logic, and The setup is based on statistical methods: First, a database of heat-processed labeled images, manually categorized into three classes—clear, moderately degraded, and severely degraded—is established; then, the characteristics of each image in the database are calculated. Values; finally, plot the three classes of samples. The threshold is determined by finding the boundary point where the overlap between class distributions is minimized in the histogram of value distributions; for example... Can be set to clear class The 95th percentile of the distribution and the moderately degenerate class The mean of the 5th quantile of the distribution. The setting of the threshold also follows this principle; this threshold calibration method ensures that the setting of the decision threshold has a statistical basis and technical reproducibility.

[0024] Example 5 When a clear marker signal is generated, the hierarchical detection unit uses a standard template matching algorithm to process the enhanced image and obtain the marker recognition result; When a moderately degraded signal is generated, the hierarchical detection unit uses a lightweight convolutional neural network model to process the enhanced image and obtain the label recognition result; When a severely degraded signal is generated, the processing procedure of the hierarchical detection unit is as follows: a generative adversarial network is used to repair the enhanced image to obtain a repaired image; a lightweight convolutional neural network model is used to process the repaired image to obtain the label recognition result. In this specific embodiment, after receiving the signal generated by the aforementioned decision logic, the hierarchical detection unit will invoke the corresponding processing strategy, thereby achieving efficient utilization of computing resources while ensuring high accuracy. When generating a clear marker signal, for cases with high image quality, the unit will invoke a standard template matching algorithm that is fast in computation and low in resource consumption. The system slides up a predefined template and calculates similarity to quickly and accurately complete the recognition. Correspondingly, when the judgment label has undergone moderate degradation, its recognition robustness may decrease. In this case, the unit calls a pre-trained lightweight convolutional neural network model to... After processing, the model has stronger feature extraction and generalization capabilities, and can effectively identify markers with deformation and missing information. In cases of severe label degradation, this unit will initiate a two-stage processing flow that includes restoration and recognition; this flow does not begin with direct recognition, but instead uses a generative adversarial network to enhance the image. The repair process leverages its advantages in image distribution learning to repair and reconstruct broken stroke structures, outputting a repaired image. This restored image The data is then fed into the aforementioned lightweight convolutional neural network model for final recognition. This strategy of repairing first and then recognizing enables the system to obtain reliable recognition results even when faced with severely degraded markers that are difficult to handle by traditional methods, demonstrating excellent detection robustness.

[0025] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A vision-based detection system for detecting heat treatment markings on wooden pallets, characterized in that, include: The image enhancement unit is used to perform wood grain suppression and contrast enhancement processing on the region of interest within the acquired wooden pallet image to obtain an enhanced image; A degradation assessment unit is used to perform frequency domain energy distribution analysis on the enhanced image to obtain a surface degradation index; The hierarchical detection unit is used to perform hierarchical detection processing on the enhanced image based on the surface degradation index to obtain the marker recognition result.

2. The vision-based inspection system for detecting heat treatment marks on wooden pallets according to claim 1, characterized in that, The wood grain suppression and contrast enhancement process is as follows: Calculate the gradient information of each pixel in the region of interest; calculate the wood grain directionality suppression weight based on the gradient information and the degree of deviation from the preset orthogonal direction; obtain the local contrast information of the region of interest; and generate an enhanced image by combining the wood grain directionality suppression weight and the local contrast information.

3. The vision-based inspection system for detecting heat treatment marks on wooden pallets according to claim 1, characterized in that, The frequency domain energy distribution analysis process is as follows: A two-dimensional discrete Fourier transform is performed on the enhanced image to obtain the centered spectrum; the energy of the predefined high-frequency region in the centered spectrum and the total energy of the full spectrum region are calculated; based on the energy of the high-frequency region and the total energy, the surface degradation index is calculated.

4. The vision-based inspection system for detecting heat treatment marks on wooden pallets according to claim 3, characterized in that, The predefined high-frequency region is set according to the higher frequency range outside the characteristic frequency band of the corresponding mark stroke in the characteristic spectrum of the standard heat-processed mark image.

5. A vision-based detection system for detecting heat treatment marks on wooden pallets according to claim 1, characterized in that, The graded detection process is as follows: The surface degradation index is compared and analyzed with preset low degradation thresholds and preset high degradation thresholds; when the surface degradation index is less than the preset low degradation threshold, a clear marking signal is generated; When the surface degradation index is greater than or equal to the preset low degradation threshold and less than the preset high degradation threshold, a moderate degradation signal is generated; A severe degradation signal is generated when the surface degradation index is greater than or equal to the preset high degradation threshold.

6. A vision-based detection system for detecting heat treatment marks on wooden pallets according to claim 5, characterized in that, When the clear marker signal is generated, the hierarchical detection unit uses a standard template matching algorithm to process the enhanced image to obtain the marker recognition result.

7. A vision-based detection system for detecting heat treatment marks on wooden pallets according to claim 5, characterized in that, When the moderate degradation signal is generated, the hierarchical detection unit uses a lightweight convolutional neural network model to process the enhanced image to obtain the marker recognition result.

8. A vision-based detection system for detecting heat treatment marks on wooden pallets according to claim 5, characterized in that, When the severe degradation signal is generated, the processing procedure of the graded detection unit is as follows: Generative adversarial networks are used to repair the enhanced image, resulting in a repaired image; a lightweight convolutional neural network model is then used to process the repaired image to obtain the label recognition result.

Images