Method and system for detecting reticular cementing quality based on AI vision

Abstract

The invention relates to an AI vision-based reticular cementation quality detection method and system, and the method comprises the steps: obtaining target temperature distribution data and target surface image data corresponding to reticular cementation, determining a target abnormal feature set corresponding to the target surface image data, carrying out the classification processing of the target abnormal feature set, and obtaining a target abnormal feature set; the method comprises the following steps of: obtaining an abnormal classification result corresponding to each region in reticular cementation, extracting a local temperature gradient value of each region from target temperature distribution data, and obtaining abnormal correlation data corresponding to the reticular cementation based on the abnormal classification result corresponding to each region and the local temperature gradient value. And extracting temperature gradient data and crack texture data corresponding to each region from the abnormal correlation data, and obtaining a quality detection result corresponding to the mesh cementation based on the temperature gradient data and the crack texture data corresponding to each region. By means of the method, the defects of the net-shaped cement can be rapidly determined, and the accuracy of quality detection of the net-shaped cement is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of cementing quality detection, and particularly relates to a mesh cementing quality detection method and system based on AI vision. BACKGROUND

[0002] In the technical field of cementing, a sintering process is usually used to process mesh cementing materials. The sintering process combines material particles of the mesh cementing materials through high temperature to obtain a cementing product. However, there are defective products in the cementing product, such as cracks and protrusions, which seriously affect the quality of the cementing product.

[0003] Since the change of the temperature field is a core factor affecting the mesh cementing materials in the sintering process, the change of the temperature field can cause stress concentration in the mesh cementing materials, thereby causing cracks or structural defects. Therefore, in order to ensure the sintering quality of the mesh cementing materials, AI vision is introduced to monitor the change of the mesh cementing materials in real time. However, it is a technical problem to monitor and analyze the change trend of the temperature field in real time and accurately. The non-uniformity of the temperature field is difficult to be fully captured by a sensor, and the change of the temperature field directly affects the micro topography of the surface of the mesh cementing materials. The change trend of the temperature field captured by the sensor and the image of the mesh cementing materials captured by the AI vision cannot be effectively fused, so that the quality defects caused by temperature abnormalities cannot be quickly identified. SUMMARY

[0004] The application provides a mesh cementing quality detection method and system based on AI vision, which realizes effective fusion of the change trend of the temperature field and the image of the mesh cementing materials captured by the AI vision, thereby improving the accuracy of identifying the quality defects of the mesh cementing.

[0005] In a first aspect, the application provides a mesh cementing quality detection method based on AI vision, which comprises the following steps: obtaining target temperature distribution data and target surface image data corresponding to mesh cementing; determining a target abnormal feature set corresponding to the target surface image data, classifying the target abnormal feature set to obtain abnormal classification results corresponding to each region of the mesh cementing respectively; extracting local temperature gradient values of the regions from the target temperature distribution data, and obtaining abnormal correlation data corresponding to the mesh cementing based on the abnormal classification results corresponding to each region respectively and the local temperature gradient values; extracting temperature gradient data and crack texture data corresponding to each region respectively from the abnormal correlation data, and obtaining quality detection results corresponding to the mesh cementing based on the temperature gradient data and the crack texture data corresponding to each region respectively.

[0006] In a possible design, the obtaining the target temperature distribution data corresponding to the network cementation comprises: obtaining time-series temperature data of the network cementation; determining all temperature gradient values corresponding to each region respectively from the time-series temperature data, and calculating a standard deviation value corresponding to each region respectively, wherein the standard deviation value is proportional to a temperature change degree; if the standard deviation value is greater than a preset standard deviation value, determining a corresponding region as a target region, and determining temperature data corresponding to the target region as the target temperature distribution data.

[0007] In a possible design, the obtaining the target surface image data corresponding to the network cementation comprises: obtaining target image data of the network cementation, and obtaining a temperature abnormal region corresponding to the target temperature distribution data; determining a gray value corresponding to the temperature abnormal region in the target image data, and if the gray value is greater than a preset gray value, determining an enlarged image data corresponding to the temperature abnormal region; determining temperature distribution data corresponding to the temperature abnormal region from the target temperature distribution data, and generating initial surface image data based on the enlarged image data and the temperature distribution data; performing correlation analysis on the initial surface image data to obtain the target surface image data.

[0008] In a possible design, the performing correlation analysis on the initial surface image data to obtain the target surface image data comprises: performing decomposition on the initial surface image data to obtain crack image data and cementation temperature distribution data; determining an edge gray value of a crack edge region in the crack image data, and if the edge gray value is lower than a preset edge gray value, performing enhancement processing on the crack edge region to obtain crack detail image data; generating the target surface image data based on the crack detail image data and the cementation temperature distribution data.

[0009] In a possible design, the determining the target abnormal feature set corresponding to the target surface image data comprises: performing multi-scale decomposition on the target surface image data to obtain component feature data of different scales, and fusing the component feature data of different scales to obtain surface image data; determine a peak signal-to-noise ratio corresponding to the surface image data, and if the peak signal-to-noise ratio is greater than a preset peak signal-to-noise ratio, perform quality evaluation on the surface image data to obtain image quality evaluation data, wherein the peak signal-to-noise ratio represents a distortion degree of the surface image data; perform crack identification processing on the image quality evaluation data to obtain an initial crack distribution map; perform crack feature extraction on the initial crack distribution map to obtain a global crack feature set, and determine the global crack feature as the target anomaly feature set.

[0010] In a possible design, the classification processing on the target anomaly feature set to obtain respective anomaly classification results corresponding to each of the regions in the network cementation includes: perform dimension reduction processing on a target anomaly feature in the target anomaly feature set to obtain a dimension-reduced feature set; perform normalization processing on a dimension-reduced feature in the dimension-reduced feature set to obtain a stable feature set; perform classification processing on a stable feature in the stable feature set based on a preset fully connected layer to obtain the respective anomaly classification results corresponding to each of the regions.

[0011] In a possible design, the obtaining of the anomaly correlation data corresponding to the network cementation based on the respective anomaly classification results corresponding to each of the regions and the local temperature gradient value includes: determine a surface feature distribution attribute and an internal anomaly feature distribution attribute from the respective anomaly classification results corresponding to each of the regions, wherein the surface feature distribution attribute represents a morphology and distribution rule of cracks on a surface of the network cementation, and the internal anomaly feature distribution attribute represents a stress concentration area of cracks in the network cementation; perform fusion processing on the surface feature distribution attribute and the local temperature gradient value corresponding to each of the regions to obtain fusion feature data; determine a feature variance corresponding to the fusion feature data, and if the feature variance is greater than a preset feature variance, perform classification processing on the fusion feature data to obtain the anomaly correlation data corresponding to the network cementation.

[0012] In a possible design, the obtaining of the quality detection result of the network cementation based on the respective temperature gradient data and crack texture data corresponding to each of the regions includes: extract a temperature distribution gradient value corresponding to each of the regions from the temperature gradient data, and extract a crack texture feature corresponding to each of the regions from the crack texture data; determine a spatial distribution overlap degree between the temperature gradient value corresponding to each of the regions and the crack texture feature; if the spatial distribution overlap degree is greater than a preset spatial distribution overlap degree, determine the corresponding region as a high correlation region, and extract crack distribution information from the high correlation region; extract surface anomaly data from the anomaly correlation data, and determine a crack distribution position and a crack type corresponding to the network cement based on the surface anomaly data and the crack distribution information; generate a quality detection result corresponding to the network cement based on the crack distribution position and the crack type.

[0013] In a second aspect, the present application provides a network cement quality detection system based on AI vision, which comprises: an acquisition module configured to acquire target temperature distribution data and target surface image data corresponding to network cement; a classification module configured to determine a target anomaly feature set corresponding to the target surface image data, and perform classification processing on the target anomaly feature set to obtain anomaly classification results corresponding to each of the regions in the network cement; an association module configured to extract local temperature gradient values of the regions from the target temperature distribution data, and obtain anomaly correlation data corresponding to the network cement based on the anomaly classification results corresponding to each of the regions and the local temperature gradient values; a detection module configured to extract temperature gradient data and crack texture data corresponding to each of the regions from the anomaly correlation data, and obtain a quality detection result corresponding to the network cement based on the temperature gradient data and the crack texture data corresponding to each of the regions.

[0014] In a possible design, the acquisition module is specifically configured to acquire time series temperature data of the network cement, determine all temperature gradient values corresponding to each of the regions from the time series temperature data, calculate standard deviation values corresponding to each of the regions, wherein the standard deviation values are proportional to the degree of temperature change, and if the standard deviation value is greater than a preset standard deviation value, determine the corresponding region as a target region, and determine temperature data corresponding to the target region as the target temperature distribution data.

[0015] In a possible design, the acquisition module is further configured to acquire target image data of the network cementation, acquire a temperature anomaly area corresponding to the target temperature distribution data, determine a gray value corresponding to the temperature anomaly area in the target image data, determine magnified image data corresponding to the temperature anomaly area if the gray value is greater than a preset gray value, determine temperature distribution data corresponding to the temperature anomaly area from the target temperature distribution data, generate initial surface image data based on the magnified image data and the temperature distribution data, perform correlation analysis on the initial surface image data, and obtain the target surface image data.

[0016] In a possible design, the acquisition module is further configured to decompose the initial surface image data to obtain crack image data and cementation temperature distribution data, determine an edge gray value of a crack edge area in the crack image data, perform enhancement processing on the crack edge area to obtain crack detail image data if the edge gray value is lower than a preset edge gray value, and generate the target surface image data based on the crack detail image data and the cementation temperature distribution data.

[0017] In a possible design, the classification module is specifically configured to perform multi-scale decomposition on the target surface image data to obtain component feature data of different scales, fuse the component feature data of different scales to obtain surface image data, determine a peak signal-to-noise ratio corresponding to the surface image data, perform quality evaluation on the surface image data to obtain image quality evaluation data if the peak signal-to-noise ratio is greater than a preset peak signal-to-noise ratio, wherein the peak signal-to-noise ratio represents a distortion degree of the surface image data, perform crack recognition processing on the image quality evaluation data to obtain an initial crack distribution map, perform crack feature extraction on the initial crack distribution map to obtain a global crack feature set, and determine the global crack feature set as the target anomaly feature set.

[0018] In a possible design, the classification module is further configured to perform dimension reduction processing on a target anomaly feature in the target anomaly feature set to obtain a dimension-reduced feature set, perform normalization processing on a dimension-reduced feature in the dimension-reduced feature set to obtain a stable feature set, and perform classification processing on a stable feature in the stable feature set based on a preset full connection layer to obtain an anomaly classification result corresponding to each of the regions.

[0019] In a possible design, the correlation module is specifically configured to determine surface feature distribution attributes and internal abnormal feature distribution attributes from the respective abnormal classification results of the respective regions, where the surface feature distribution attributes represent the morphology and distribution law of the cracks on the net-like cementation surface, and the internal abnormal feature distribution attributes represent stress concentration areas of the cracks in the net-like cementation interior, perform fusion processing on the respective surface feature distribution attributes and the local temperature gradient values of the respective regions to obtain fusion feature data, determine a feature variance corresponding to the fusion feature data, and perform classification processing on the fusion feature data if the feature variance is greater than a preset feature variance to obtain the abnormal correlation data corresponding to the net-like cementation.

[0020] In a possible design, the detection module is specifically configured to extract temperature distribution gradient values corresponding to the respective regions from the temperature gradient data, extract crack texture features corresponding to the respective regions from the crack texture data, determine a spatial distribution overlap degree between the temperature distribution gradient values and the crack texture features corresponding to the respective regions, determine a high-correlation region if the spatial distribution overlap degree is greater than a preset spatial distribution overlap degree, extract crack distribution information from the high-correlation region, extract surface abnormal data from the abnormal correlation data, determine a crack distribution position and a crack type corresponding to the net-like cementation based on the surface abnormal data and the crack distribution information, and generate a quality detection result corresponding to the net-like cementation based on the crack distribution position and the crack type.

[0021] In a third aspect, the present application provides an electronic device, comprising: a memory for storing a computer program; a processor for executing the computer program stored on the memory to implement the AI vision-based net-like cementation quality detection method steps.

[0022] In a fourth aspect, a computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the AI vision-based net-like cementation quality detection method steps.

[0023] The technical effects that can be achieved by each of the above-mentioned first to fourth aspects and each aspect are described above in the technical effects that can be achieved by the first aspect or each possible solution in the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of the AI vision-based net-like cementation quality detection method steps provided by the present application; Figure 2 A structure schematic diagram of an AI vision-based mesh cementation quality detection system provided by the present application is provided. Figure 3 A structure schematic diagram of an electronic device provided by the present application is provided. DETAILED DESCRIPTION

[0025] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings. The specific operation methods in the method embodiments can also be applied to the device embodiments or system embodiments. It should be noted that in the description of the present application, "multiple" is understood as "at least two". The association relationship of the associated objects is described, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. A and B are connected, which means that A and B are directly connected and A and B are connected through C. In addition, in the description of the present application, "first", "second", etc. are used only for the purpose of distinguishing the description, and cannot be understood as indicating or implying relative importance, nor can it be understood as indicating or implying order.

[0026] In the past technology, in order to ensure the sintering quality of the mesh cementation material, AI vision is introduced to monitor the change of the mesh cementation material in real time. However, it is a technical problem to monitor and analyze the change trend of the temperature field in real time and accurately. The non-uniformity of the temperature field is difficult to be fully captured by the sensor, and the change of the temperature field will directly affect the micro-topography of the surface of the mesh cementation material. The change trend of the temperature field captured by the sensor and the image of the mesh cementation material captured by the AI vision cannot be effectively fused, which leads to the inability to quickly identify the quality defects caused by temperature abnormalities.

[0027] In order to solve the above-mentioned problems, the present application provides an AI vision-based mesh cementation quality detection method, which realizes effective fusion between the change trend of the temperature field and the image of the mesh cementation material captured by the AI vision, so as to ensure that the quality defects caused by temperature abnormalities can be quickly identified. The method and device provided by the present application are based on the same technical concept. Since the principles of the problems solved by the method and the device are similar, the embodiments of the device and the method can be mutually referred to, and the repeated parts will not be described again.

[0028] The first embodiment of the present application will be described in detail below with reference to the drawings.

[0029] Reference Figure 1The application provides a mesh cementation quality detection method based on AI vision, which can realize effective fusion between the temperature field change trend and the image of the mesh cementation material captured by AI vision, so as to ensure that the quality defects caused by temperature abnormalities can be quickly identified, and the implementation process of the method is as follows: Step S1: Obtain target temperature distribution data and target surface image data corresponding to the mesh cementation.

[0030] Since the temperature field distribution of the mesh cementation is not uniform during the sintering process, and the temperature sensor is affected by the temperature, a spatial rectangular coordinate system needs to be established, and the spatial coordinates of the temperature sensor can be obtained based on the spatial rectangular coordinate system. The temperature sensor is arranged at the key points of the mesh cementation, and the key points are the structural stress points and the mesh structure intersection points of the mesh cementation. Here, no more description is made. The temperature sensor collects the temperature data of the mesh cementation at a set frequency, for example, at a frequency of 10 times per second.

[0031] In a possible design, since the temperature sensors at different positions are affected by different degrees of environmental interference, a weight value can be set for each temperature sensor. In the embodiment of the application, the weight value of the temperature sensor corresponding to the edge area can be set to 0.4, and the temperature sensor close to the core area of the mesh cementation is less affected by the environmental interference, and the weight value of the temperature sensor in the core area can be set to 0.6. The weight value of the temperature sensor can be adjusted according to the actual situation, and no more description is made here.

[0032] Based on the above method, the influence of environmental interference on the temperature sensor is reduced, and the accuracy of the temperature data collected by the temperature sensor is ensured.

[0033] Further, the original temperature value collected by the temperature sensor is obtained, the original temperature value corresponding to each region in the mesh cementation is weighted and averaged based on the weight value corresponding to each temperature sensor, the temperature value corresponding to each region is obtained, and the temperature deviation between the temperature value corresponding to each region and at least one original temperature value is determined. If the temperature deviation between the temperature value and the original temperature value is greater than the preset temperature deviation, the preset temperature deviation can be 3 degrees, the corresponding temperature value is determined as an abnormal temperature, and the abnormal temperature is deleted to obtain smooth temperature data.

[0034] In order to smooth the integrity of the temperature data, the missing temperature value in the smooth temperature data needs to be filled based on the linear interpolation algorithm to obtain dynamic temperature sequence data.

[0035] For example, the temperature value corresponding to 8:05:00 in a certain area of the network cementation is 24.5°C, the temperature value corresponding to 8:07:00 is 25.5°C, and if the temperature value corresponding to 8:06:00 is missing, then the temperature value corresponding to 8:06:00 is 25°C based on the linear interpolation algorithm, and 25°C is determined as the temperature value corresponding to 8:06:00.

[0036] The dynamic temperature sequence data in the embodiment of the application is shown in Table 1: Table 1 The above Table 1 records the respective space coordinates and temperature values of each area in the network cementation at T1 time. The embodiment of the application can adjust the time stamp, space coordinate and temperature value according to the actual parameters of the network cementation in the sintering process, and can also collect the dynamic temperature sequence data of one area in the network cementation, as shown in Table 2: Table 2 The above Table 2 records the dynamic temperature sequence data of one area in the network cementation, and the dynamic temperature sequence data of other areas is referred to the above Table 2, which is not described in detail here.

[0037] After determining the dynamic temperature sequence data corresponding to the network cementation, the temperature gradient value corresponding to each area can be determined based on the dynamic temperature sequence data. The temperature gradient value represents the direction and rate of temperature change in the temperature field. Since the calculation method of the temperature gradient value is known to those skilled in the art, it is not described in detail here.

[0038] Based on the dynamic temperature sequence data and the temperature gradient value corresponding to each area, time sequence temperature data is generated, which at least includes the respective space coordinates and temperature gradient values of each area. In order to intuitively reflect the temperature distribution, it is necessary to map the respective space coordinates and temperature gradient values of each area to a temperature gradient matrix based on a spatial distribution mapping algorithm. The spatial distribution mapping algorithm can be linear discriminant analysis (English full name: Linear Discriminant Analysis, abbreviated as: LDA), kernel principal component analysis (English full name: Kernel Principal Component Analysis, abbreviated as: KPCA), etc., which is not described in detail here.

[0039] To obtain a more accurate temperature distribution rule, the embodiment of the application can divide each region into a plurality of sub-regions, and obtain the temperature gradient value corresponding to each sub-region. The manner of determining the temperature gradient value of the sub-region is consistent with the manner of determining the temperature gradient value of each region, which will not be described in detail here. Based on all the temperature values of the sub-region, the standard deviation value is determined, which is proportional to the degree of temperature change. The standard deviation value can be the average of all the temperature values of the sub-region, thereby facilitating the determination of the region with irregular temperature distribution.

[0040] When the standard deviation value is greater than the preset standard deviation value, the corresponding region is determined as the target region, and the temperature data corresponding to the target region is selected from the time series temperature data, and the temperature data is determined as the target temperature distribution data.

[0041] For example: A region is divided into a1, a2, and a3, which are sub-regions of A region, and the temperature gradient values of a1, a2, and a3 are 0.3, 0.2, and 0.4 respectively. The calculated standard deviation value is 0.3, and the preset standard deviation value is 0.2. 0.3>0.2, which means that the temperature change of A region is large, so it can be determined that the temperature distribution of A region is uneven.

[0042] The preset standard deviation value in the embodiment of the application can be adjusted based on actual conditions, which will not be described in detail here.

[0043] In order to realize the quality detection of the network cementation, it is necessary to determine the target surface image data of the network cementation in the sintering process. The specific process of determining the target surface image data is as follows: The embodiment of the application is AI vision to realize the monitoring of the network cementation. The AI vision is realized by an intelligent infrared imaging device. The intelligent infrared imaging device is used to collect the original image of the network cementation. The original image is denoised. The embodiment of the application can denoise the original image based on median filtering to obtain a denoised image. The filter window can be 3x3 pixels, which can effectively remove random noise in the original image while preserving edge details. Then, the denoised image is enhanced based on histogram equalization to obtain target image data. The enhancement processing expands the gray value range from the original 0-255 to a wider range, thereby improving the structure visibility of the network cementation in the target image data.

[0044] Based on the above method, the target image data accurately reflects the texture of the network cementation and the uniformity of the network cementation.

[0045] In order to improve the detection efficiency of the quality of the network cementation, the system needs to determine the temperature abnormal area corresponding to the target temperature distribution data, and determine the gray value corresponding to the temperature abnormal area in the target image data. If the gray value is greater than the preset gray value, the magnified image data corresponding to the temperature abnormal area is determined, and the magnified image data reveals the local topographic features such as micro-cracks or pores.

[0046] In order to realize the effective fusion of the temperature distribution data and the image data, the temperature distribution data corresponding to the temperature abnormal area needs to be determined from the target temperature distribution data, and the correlation between the magnified image data and the temperature distribution data is determined based on the spatial coordinates, and the initial surface image data is generated. The initial surface image data represents the correlation characteristics of the surface of the temperature abnormal area in the network cementation and the temperature field anomaly.

[0047] The number of the above-mentioned temperature abnormal area is at least one. In the embodiment of the application, one temperature abnormal area is described, and the other temperature abnormal areas are described with reference to the above-mentioned content, which will not be described here.

[0048] In a possible design, the embodiment of the application can fuse the magnified image data and the temperature distribution data based on the weighted average algorithm. The weight value of the magnified image data can be set to 0.6, and the weight value of the temperature distribution data can be set to 0.4 to generate the initial surface image data.

[0049] The weight value of the magnified image data and the weight value of the temperature distribution data in the embodiment of the application can be adjusted according to the actual situation, which will not be described here.

[0050] After the initial surface image data is determined, the initial surface image data is analyzed to obtain the target surface image data. The specific process of the correlation analysis of the initial surface image data is as follows: The initial surface image data is decomposed to obtain high-frequency components and low-frequency components. The decomposition method can be wavelet decomposition. The high-frequency components correspond to the crack image data, and the low-frequency components correspond to the cementation temperature distribution data. The edge gray value of the crack edge region in the crack image data is determined. If the edge gray value is lower than the preset edge gray value, the crack edge region is enhanced to obtain the crack detail image data. The crack detail image data and the cementation temperature distribution data are fused based on the weighted average algorithm to generate the target surface image data. The target surface image data is beneficial to analyze the micro-topography and potential defects of the network cementation surface.

[0051] In a possible design, the embodiment of the application can also denoise the crack image data based on the threshold shrinkage method. The specific denoising process is as follows: Determine the gray scale standard deviation corresponding to the crack image data, the greater the gray scale standard deviation, the higher the contrast of the crack image data, set the threshold value as 1.5 times of the gray scale standard deviation, set the pixel value of the crack image data less than the set threshold value to zero, and compress the pixel value greater than the set threshold value, so that the noise points in the crack image data are effectively suppressed, and the pixels of the crack edge in the crack image data are retained, so that the micro features in the network cementation can be highlighted.

[0052] In the embodiments of the present application, the gray scale standard deviation can be calculated based on ImageJ, and the above-mentioned 1.5 can be adjusted according to the actual situation, which will not be described here.

[0053] The denoised crack image data is input into a non-local mean filter (English full name: Non-Local Means, abbreviated as: NLM) for enhancement processing to obtain enhanced crack image data.

[0054] After determining the denoised and enhanced crack image data, the crack detail image data corresponding to the denoised and enhanced crack image data is determined based on the above-described method, and the target surface image data is generated based on the crack detail image data and the cementation temperature distribution data.

[0055] Based on the above method, the multi-level processing method ensures the accuracy of the target surface image data, which is beneficial to improve the accuracy of the quality detection of the network cementation.

[0056] Step S2: Determine the target abnormal feature set corresponding to the target surface image data, and classify the target abnormal feature set to obtain the abnormal classification result corresponding to each region in the network cementation.

[0057] Based on the wavelet basis function, the target surface image data is multi-scale decomposed to obtain component feature data of different scales, the component feature data includes high-frequency component data and low-frequency component data, the high-frequency component data is used to capture fine features such as cracks, and the low-frequency component data reflects the overall surface morphology of the network cementation.

[0058] In order to fuse the component feature data of different scales, it is necessary to fuse the component feature data of different scales based on the weighted average algorithm to obtain the surface image data, since the weighted average algorithm is a known technology to those skilled in the art, therefore, it will not be described here.

[0059] Determine the peak signal-to-noise ratio corresponding to the surface image data, the peak signal-to-noise ratio represents the distortion degree of the surface image data, the higher the distortion degree, the smaller the peak signal-to-noise ratio, if the peak signal-to-noise ratio is greater than the preset peak signal-to-noise ratio, the quality of the surface image data is evaluated based on the gray scale distribution and noise residue to obtain image quality evaluation data.

[0060] Further, the embodiment of the present application can extract the geometric shape of the crack based on the edge detection method, and identify the crack in the image quality evaluation data based on the geometric shape of the crack and the gray difference of the adjacent area, to obtain an initial crack distribution map, which can be presented in the form of a heat map.

[0061] After determining the initial crack distribution map, since the distribution of the crack presents an irregular network shape, the initial crack distribution map is subjected to crack feature extraction based on a residual connection network layer to obtain a global crack feature set, and the global crack feature set is determined as a target abnormal feature set, so that the correlation between the local texture and the overall extension trend of the crack is obtained.

[0062] In order to reduce the amount of data and thus ensure the efficiency of the network cementation quality detection, the target abnormal features in the target abnormal feature set need to be subjected to dimension reduction processing to obtain a dimension reduction feature set, and the dimension reduction mode can be realized based on a pooling layer, which will not be described in detail here.

[0063] For example, the pooling layer adopts a 2x2 maximum pooling operation to reduce the feature map size corresponding to the target abnormal features from 256x256 to 128x128.

[0064] The dimension reduction features in the dimension reduction feature set are subjected to normalization processing to obtain a stable feature set. In order to classify the cracks, the stable features in the stable feature set need to be subjected to classification processing based on a preset full connection layer to obtain abnormal classification results corresponding to each region respectively, which at least include the crack degree and the crack probability corresponding to the crack degree.

[0065] For example, the probability of slight crack is 0.7, the probability of severe crack is 0.2, and the probability of no crack is 0.1.

[0066] Through the above method, the abnormal classification results are obtained based on the fine texture of the crack and the overall distribution pattern of the crack, so as to improve the recognition ability of complex cracks and ensure the accuracy of the quality detection of the network cementation.

[0067] Step S3: Extracting local temperature gradient values of each region from the target temperature distribution data, and obtaining abnormal correlation data corresponding to the network cementation based on the abnormal classification results corresponding to each region respectively and the local temperature gradient values.

[0068] The local temperature gradient values of each region are extracted from the target temperature distribution data, and the each region is a temperature abnormal region. In order to realize the quality detection of the network cementation, the surface feature distribution attribute and the internal abnormal feature distribution attribute need to be determined from the abnormal classification results, the surface feature distribution attribute represents the shape and distribution law of the crack on the surface of the network cementation, and the internal abnormal feature distribution attribute represents the stress concentration area of the crack in the network cementation.

[0069] The surface feature distribution attribute and the internal abnormal feature distribution attribute are associated to determine an initial correlation attribute set corresponding to the network cementation, the initial correlation attribute set including an association value between the surface feature distribution attribute and the internal abnormal feature distribution attribute corresponding to each region, the association value being a parameter between 0 and 1, and the association value being proportional to the association strength.

[0070] The embodiment of the application can determine the association value between the surface feature distribution attribute and the internal abnormal feature distribution attribute based on the Pearson correlation coefficient method. Since the Pearson correlation coefficient method is a known technology for those skilled in the art, it will not be described in detail here.

[0071] In a possible design, the embodiment of the application can determine a surface crack density value corresponding to the surface feature distribution attribute in the abnormal classification result, and determine an abnormal region proportion value corresponding to the internal abnormal feature distribution attribute, so as to more intuitively display the defect region of the network cementation based on the surface crack density value and the abnormal region proportion value.

[0072] The surface crack density value can be a ratio between the total length of the crack and the area of the observation section, or the number of cracks per unit length. The abnormal region proportion value can be a ratio between the number of temperature abnormal regions and the total number of regions. Details are not described here.

[0073] The surface feature distribution attribute and the local temperature gradient value corresponding to each region are fused to obtain fusion feature data, and a feature variance corresponding to the fusion feature data is determined. If the feature variance is greater than a preset feature variance, the fusion feature data is classified to obtain abnormal correlation data corresponding to the network cementation.

[0074] For example, the variance of the feature distribution of the fusion feature data is 0.2, the preset feature variance is 0.3, 0.2 is less than 0.3, and the preset condition is met. The fusion feature data is processed by the full connection layer to output the abnormal correlation data. The abnormal correlation data shows that the correlation probability between the surface crack and the internal abnormality is 0.75.

[0075] Through the above method, the surface feature distribution attribute and the internal abnormal feature distribution attribute are fused to realize the association between the surface crack and the internal temperature, so that the crack feature determined in the sintering process of the network cementation is more accurate, thereby improving the accuracy of the quality detection of the network cementation.

[0076] Step S4: Extracting temperature gradient data and crack texture data corresponding to each region from the abnormal correlation data, and obtaining a quality detection result corresponding to the network cementation based on the temperature gradient data and the crack texture data corresponding to each region.

[0077] The temperature gradient data corresponding to each region is extracted from the abnormal correlation data, the temperature gradient data reflecting the temperature change rate of the network cement in different regions, the crack texture data describing the morphology and distribution characteristics of the cracks on the surface of the network cement, the temperature distribution gradient value corresponding to each region is extracted from the temperature gradient data, and the crack texture feature corresponding to each region is extracted from the crack texture data, the crack texture feature including a crack length, a crack width, and a crack density, a spatial distribution overlap degree between the temperature distribution gradient value corresponding to each region and the crack texture feature is determined, and a correlation strength distribution diagram corresponding to the network cement is obtained, which can intuitively show the spatial correspondence between the cracks and the temperature distribution gradient value.

[0078] In the embodiments of the present application, each region corresponds to only one temperature distribution gradient value, the temperature distribution gradient value can be the same as the standard deviation value, and the spatial distribution overlap degree between the temperature distribution gradient value and the crack texture feature is calculated through topological operation, element-level Boolean operation, etc., which will not be described in detail here.

[0079] Further, the temperature distribution gradient value can be classified by gradient to obtain different temperature gradients, and the crack texture data can be locally segmented to obtain local segmented cracks, and the correlation strength distribution diagram between the different temperature gradients and the corresponding local segmented cracks can be determined based on the above method.

[0080] For example, [0, 2) degrees per centimeter is a low gradient, [2, 5) degrees per centimeter is a medium gradient, and 5 degrees per centimeter or more is a high gradient, and the gradient classification can clearly distinguish different temperature change regions.

[0081] If the spatial distribution overlap degree is greater than a preset spatial distribution overlap degree, which can be 0.8, the corresponding region is determined as a high correlation region, crack distribution information is extracted from the high correlation region, surface anomaly data is extracted from the abnormal correlation data, the crack distribution position and the crack type corresponding to the network cement are determined based on the surface anomaly data and the crack distribution information, and the quality detection result corresponding to the network cement is generated based on the crack distribution position and the crack type.

[0082] Through the above method, the target temperature distribution data and the target surface image data of the network cement in the sintering process are combined, the effective fusion of the surface defects and the internal defects of the network cement is realized, the correlation detection based on the temperature gradient data and the crack texture data is realized, the quality detection efficiency and the accuracy of the network cement are improved, and the precise positioning of the cracks is realized.

[0083] Based on the same inventive concept, the application also provides an AI vision-based mesh cementation quality detection system, and the encryption transmission device is used to realize the function of an AI vision-based mesh cementation quality detection method. Figure 2 The device comprises: An acquisition module 201 is configured to acquire target temperature distribution data and target surface image data corresponding to mesh cementation. A classification module 202 is configured to determine a target abnormal feature set corresponding to the target surface image data, perform classification processing on the target abnormal feature set, and obtain abnormal classification results corresponding to each region in the mesh cementation. An association module 203 is configured to extract local temperature gradient values of the regions from the target temperature distribution data, and obtain abnormal correlation data corresponding to the mesh cementation based on the abnormal classification results corresponding to each region and the local temperature gradient values. A detection module 204 is configured to extract temperature gradient data and crack texture data corresponding to each region from the abnormal correlation data, and obtain quality detection results corresponding to the mesh cementation based on the temperature gradient data and the crack texture data corresponding to each region.

[0084] In a possible design, the acquisition module 201 is specifically configured to acquire time series temperature data of the mesh cementation, determine all temperature gradient values corresponding to each region from the time series temperature data, calculate standard deviation values corresponding to each region, wherein the standard deviation values are proportional to the degree of temperature change, and if the standard deviation value is greater than a preset standard deviation value, the corresponding region is determined as a target region, and the temperature data corresponding to the target region is determined as the target temperature distribution data.

[0085] In a possible design, the acquisition module 201 is further configured to acquire target image data of the mesh cementation, acquire a temperature abnormal region corresponding to the target temperature distribution data, determine a gray value corresponding to the temperature abnormal region in the target image data, if the gray value is greater than a preset gray value, determine enlarged image data corresponding to the temperature abnormal region, determine temperature distribution data corresponding to the temperature abnormal region from the target temperature distribution data, generate initial surface image data based on the enlarged image data and the temperature distribution data, and perform association analysis on the initial surface image data to obtain the target surface image data.

[0086] In a possible design, the acquisition module 201 is further configured to perform decomposition on the initial surface image data to obtain crack image data and cementation temperature distribution data, determine an edge gray value of a crack edge region in the crack image data, perform enhancement processing on the crack edge region to obtain crack detail image data if the edge gray value is lower than a preset edge gray value, and generate the target surface image data based on the crack detail image data and the cementation temperature distribution data.

[0087] In a possible design, the classification module 202 is specifically configured to perform multi-scale decomposition on the target surface image data to obtain component feature data of different scales, fuse the component feature data of different scales to obtain surface image data, determine a peak signal-to-noise ratio corresponding to the surface image data, perform quality evaluation on the surface image data to obtain image quality evaluation data if the peak signal-to-noise ratio is greater than a preset peak signal-to-noise ratio, wherein the peak signal-to-noise ratio represents a distortion degree of the surface image data, perform crack recognition processing on the image quality evaluation data to obtain an initial crack distribution map, perform crack feature extraction on the initial crack distribution map to obtain a global crack feature set, and determine the global crack feature set as the target abnormal feature set.

[0088] In a possible design, the classification module 202 is further configured to perform dimension reduction processing on a target abnormal feature in the target abnormal feature set to obtain a dimension-reduced feature set, perform normalization processing on a dimension-reduced feature in the dimension-reduced feature set to obtain a stable feature set, and perform classification processing on a stable feature in the stable feature set based on a preset full connection layer to obtain an abnormal classification result corresponding to each of the regions.

[0089] In a possible design, the correlation module 203 is specifically configured to determine a surface feature distribution attribute and an internal abnormal feature distribution attribute from the abnormal classification result corresponding to each of the regions, wherein the surface feature distribution attribute represents a morphology and distribution rule of a crack on the netted cementation surface, and the internal abnormal feature distribution attribute represents a stress concentration area of a crack in the netted cementation interior, perform fusion processing on the surface feature distribution attribute and the local temperature gradient value corresponding to each of the regions to obtain fusion feature data, determine a feature variance corresponding to the fusion feature data, and perform classification processing on the fusion feature data to obtain abnormal correlation data corresponding to the netted cementation if the feature variance is greater than a preset feature variance.

[0090] In a possible design, the detection module 204 is specifically configured to extract the respective temperature distribution gradient value of each region from the temperature gradient data, extract the respective crack texture feature of each region from the crack texture data, determine a spatial distribution overlap between the respective temperature distribution gradient value and the crack texture feature of each region, determine a corresponding region as a high correlation region if the spatial distribution overlap is greater than a preset spatial distribution overlap, extract crack distribution information from the high correlation region, extract surface anomaly data from the abnormal correlation data, determine a crack distribution position and a crack type of the network cementation based on the surface anomaly data and the crack distribution information, and generate a quality detection result of the network cementation based on the crack distribution position and the crack type.

[0091] Based on the same inventive concept, the embodiments of the present application further provide an electronic device, which can implement the functions of the foregoing AI vision-based network cementation quality detection system, refer to Figure 3 , and the electronic device comprises: at least one processor 301 and a memory 303 connected with the at least one processor 301, and the embodiments of the present application do not limit the specific connection medium between the processor 301 and the memory 303, Figure 3 in which the processor 301 and the memory 303 are connected through a bus 300. The bus 300 is represented by a thick line in Figure 3 , and the connection modes between other components are only schematically illustrated and are not limited. The bus 300 can be divided into an address bus, a data bus, a control bus, etc., and for the convenience of representation, Figure 3 in the foregoing embodiments, only one thick line is used to represent the bus, but it does not mean that there is only one bus or only one type of bus. Alternatively, the processor 301 can also be referred to as a controller, and the name is not limited.

[0092] In the embodiments of the present application, the memory 303 stores instructions executable by the at least one processor 301, and the at least one processor 301 can execute the foregoing AI vision-based network cementation quality detection method by executing the instructions stored in the memory 303. The processor 301 can implement the functions of various modules in the system as shown in Figure 2 .

[0093] Among them, the processor 301 is the control center of the device, which can connect all parts of the control device through various interfaces and lines, and realize the functions and processing data of the system by running or executing the instructions stored in the memory 303 and calling the data stored in the memory 303, so as to monitor the whole system.

[0094] In a possible design, the processor 301 can include one or more processing units, and the processor 301 can integrate an application processor and a modem processor, where the application processor mainly processes operating systems, user interfaces, and application programs, and the modem processor mainly processes wireless communication. It can be understood that the foregoing modem processor can also not be integrated into the processor 301. In some embodiments, the processor 301 and the memory 303 can be implemented on the same chip, and in some embodiments, they can also be respectively implemented on independent chips.

[0095] The processor 301 can be a general-purpose processor, for example, a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit, a field programmable gate array, or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the AI vision-based mesh cement quality detection method disclosed in the embodiments of the present application can be directly embodied as execution completed by a hardware processor, or executed by a combination of hardware and software modules in the processor.

[0096] The memory 303 can be used to store non-volatile software programs, non-volatile computer executable programs, and modules as a non-volatile computer readable storage medium. The memory 303 can include at least one type of storage medium, for example, can include a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (RAM), a static random access memory (SRAM), a programmable read-only memory (PROM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic memory, a magnetic disk, an optical disk, and the like. The memory 303 is any other medium capable of carrying or storing desired program codes in the form of instructions or data structures and capable of being accessed by a computer, but is not limited thereto. The memory 303 in the embodiments of the present application can also be a circuit or any other device capable of realizing a storage function, for storing program instructions and / or data.

[0097] By designing and programming the processor 301, the code corresponding to the AI vision-based mesh cement quality detection method introduced in the foregoing embodiments can be solidified into the chip, so that the chip can executeFigure 1 An AI vision-based mesh cementation quality detection step of the illustrated embodiment. How to design and program the processor 301 is a technology known to those skilled in the art, which will not be repeated here.

[0098] Based on the same inventive concept, the embodiments of the present application also provide a storage medium storing computer instructions, which, when running on a computer, cause the computer to execute the foregoing AI vision-based mesh cementation quality detection method.

[0099] In some possible implementation manners, various aspects of the AI vision-based mesh cementation quality detection method provided by the present application can also be implemented in the form of a program product, which includes program codes for causing the control device to execute the steps in the AI vision-based mesh cementation quality detection method according to various exemplary embodiments of the present application described above in the specification when the program product runs on the device.

[0100] Those skilled in the art should understand that the embodiments of the present application can be provided in the form of a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0101] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as a combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0102] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The device that implements the functions specified in one or more flows and / or blocks.

[0103] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or one block or multiple blocks. Figure 1 the steps of the functions specified in the flowchart or multiple flows and / or one block or multiple blocks.

[0104] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and variations.

Claims

1. A method for detecting the quality of mesh bonding based on AI vision, characterized in that: include: Obtaining target temperature distribution data and target surface image data corresponding to the network bonding; Determining a target abnormal feature set corresponding to the target surface image data, and performing classification processing on the target abnormal feature set to obtain abnormal classification results corresponding to respective regions in the network cementation; Extracting the local temperature gradient value of each region from the target temperature distribution data, and obtaining abnormal correlation data corresponding to the network cementation based on the abnormal classification results and the local temperature gradient values ​​corresponding to each region; The temperature gradient data and crack texture data corresponding to each of the regions are extracted from the abnormal correlation data, and the quality inspection results corresponding to the network bonding are obtained based on the temperature gradient data and the crack texture data corresponding to each of the regions.

2. The method according to claim 1, wherein The obtaining of target temperature distribution data corresponding to the network bonding includes: Acquiring time series temperature data of the network cementation; Determining all temperature gradient values ​​corresponding to each region from the time series temperature data, and calculating a standard deviation value corresponding to each region, wherein the standard deviation value is proportional to the degree of temperature change; If the standard deviation value is greater than the preset standard deviation value, the corresponding area is determined as the target area, and the temperature data corresponding to the target area is determined as the target temperature distribution data.

3. The method according to claim 1, wherein The obtaining of target surface image data corresponding to the networked cementation includes: Acquiring target image data of the networked cementation and acquiring temperature abnormality areas corresponding to the target temperature distribution data; determining a grayscale value corresponding to the temperature abnormality region in the target image data, and if the grayscale value is greater than a preset grayscale value, determining amplified image data corresponding to the temperature abnormality region; determining temperature distribution data corresponding to the temperature abnormality region from the target temperature distribution data, and generating initial surface image data based on the magnified image data and the temperature distribution data; Performing correlation analysis on the initial surface image data to obtain the target surface image data.

4. The method according to claim 3, wherein The performing correlation analysis on the initial surface image data to obtain the target surface image data includes: Decomposing the initial surface image data to obtain crack image data and bonding temperature distribution data; determining an edge grayscale value of a crack edge region in the crack image data, and if the edge grayscale value is lower than a preset edge grayscale value, performing enhancement processing on the crack edge region to obtain crack detail image data; The target surface image data is generated based on the crack detail image data and the bonding temperature distribution data.

5. The method according to claim 1, wherein Determining the target abnormality feature set corresponding to the target surface image data includes: Performing multi-scale decomposition on the target surface image data to obtain component feature data of different scales, and fusing the component feature data of different scales to obtain surface image data; determining a peak signal-to-noise ratio corresponding to the surface image data, and if the peak signal-to-noise ratio is greater than a preset peak signal-to-noise ratio, performing a quality assessment on the surface image data to obtain image quality assessment data, wherein the peak signal-to-noise ratio represents a degree of distortion of the surface image data; performing crack identification processing on the image quality assessment data to obtain an initial crack distribution map; Crack features are extracted from the initial crack distribution map to obtain a global crack feature set, and the global crack features are determined as the target abnormal feature set.

6. The method according to claim 1, wherein The classification process of the target abnormal feature set to obtain the abnormal classification results corresponding to each area in the network cementation includes: Performing dimensionality reduction processing on the target abnormal features in the target abnormal feature set to obtain a reduced dimensionality feature set; Normalizing the dimensionality reduction features in the dimensionality reduction feature set to obtain a stable feature set; Based on a preset fully connected layer, the stable features in the stable feature set are classified and processed to obtain abnormal classification results corresponding to each of the regions.

7. The method according to claim 1, wherein The abnormality correlation data corresponding to the network cementation is obtained based on the abnormality classification results and the local temperature gradient values ​​corresponding to each of the regions, including: Determining a surface feature distribution attribute and an internal abnormal feature distribution attribute from the abnormality classification results corresponding to each of the regions, wherein the surface feature distribution attribute characterizes the morphology and distribution pattern of the cracks on the surface of the network cementation, and the internal abnormal feature distribution attribute characterizes the stress concentration area of ​​the cracks inside the network cementation; fusing the surface feature distribution attributes and the local temperature gradient values ​​corresponding to each of the regions to obtain fused feature data; Determine the feature variance corresponding to the fused feature data; if the feature variance is greater than a preset feature variance, perform classification processing on the fused feature data to obtain abnormal correlation data corresponding to the network cementation.

8. The method according to claim 1, wherein The quality inspection result of the network bonding is obtained based on the temperature gradient data and the crack texture data corresponding to each of the regions, including: Extracting the temperature distribution gradient value corresponding to each of the regions from the temperature gradient data, and extracting the crack texture feature corresponding to each of the regions from the crack texture data; Determining the spatial distribution overlap between the temperature distribution gradient value corresponding to each of the regions and the crack texture feature; If the spatial distribution overlap is greater than a preset spatial distribution overlap, the corresponding area is determined as a high correlation area, and crack distribution information is extracted from the high correlation area; Extracting surface abnormality data from the abnormality correlation data, and determining the crack distribution position and crack type corresponding to the network bonding based on the surface abnormality data and the crack distribution information; Based on the crack distribution positions and the crack types, a quality inspection result corresponding to the network bonding is generated.

9. A network bonding quality detection system based on AI vision, characterized in that: include: An acquisition module, used for acquiring target temperature distribution data and target surface image data corresponding to the network bonding; a classification module, configured to determine a target abnormal feature set corresponding to the target surface image data, perform classification processing on the target abnormal feature set, and obtain abnormal classification results corresponding to respective regions in the network cementation; a correlation module, configured to extract the local temperature gradient value of each region from the target temperature distribution data, and obtain abnormal correlation data corresponding to the network cementation based on the abnormal classification results and the local temperature gradient values ​​corresponding to each region; A detection module is used to extract the temperature gradient data and crack texture data corresponding to each of the regions from the abnormal correlation data, and obtain a quality detection result corresponding to the network bonding based on the temperature gradient data and the crack texture data corresponding to each of the regions.

10. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the method steps described in any one of claims 1 to 8 when executing the computer program stored in the memory.

Images