Thermal insulation pipe sampling method and system for realizing warehousing based on deep learning
Through deep learning-based methods, combining environmental parameters and image features, a sampling strategy for insulation tubes is formulated, which solves the problem of low accuracy in large-scale production of random sampling methods, and achieves more efficient insulation tube quality detection.
Patent Information
- Application Number
- CN202411044145.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-08
AI Technical Summary
The existing random sampling methods cannot effectively detect potential quality problems in large-scale production, resulting in a reduced accuracy of storage insulation pipe sampling.
A deep learning-based method is adopted to build a quality system and loss coefficient by obtaining environmental parameter data, image acquisition and feature extraction, and a sampling strategy is formulated to improve sampling accuracy.
The sampling accuracy of storage insulation pipes is improved to ensure that the sampling results are more representative of the overall product quality.
Smart Images

Figure CN118918438A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of thermal insulation pipe sampling, and in particular to a thermal insulation pipe sampling method and system for warehousing based on deep learning. Background Art
[0002] Insulation pipe is a pipe material used to maintain the temperature of the medium in the pipe or equipment stable. It is usually composed of two parts: the inner pipe and the outer insulation material. The inner pipe is made of different materials, such as steel, copper, plastic, etc., and is used to transmit liquids or gases. The outer insulation material is usually a material with lower thermal conductivity, such as polyurethane foam, glass wool, rock wool, etc. Since insulation pipes are produced and stored in batches, they need to be sampled and tested regularly to analyze the quality of the insulation pipes.
[0003] The existing sampling method for insulation pipes mainly adopts random sampling method, which is to randomly select a certain number of insulation pipe samples from the entire insulation pipe inventory for testing. However, the disadvantage of this method is that in mass production, random sampling may not be able to detect potential quality problems well, and the sampling results cannot represent the overall product quality, which in turn leads to a decrease in the accuracy of storage insulation pipe sampling. Summary of the invention
[0004] The present invention provides a method and system for sampling thermal insulation pipes in storage based on deep learning, the main purpose of which is to improve the sampling accuracy of thermal insulation pipes in storage.
[0005] To achieve the above purpose, the present invention provides a method for sampling insulation pipes for storage based on deep learning, comprising:
[0006] Obtain the storage insulation pipes to be sampled and the sampling conditions, and collect the environmental parameter data corresponding to the storage insulation pipes;
[0007] In combination with the sampling conditions, the environmental parameter data is analyzed and processed to obtain environmental variables and variable observation data, and a variable chart corresponding to the environmental variables is constructed based on the variable observation data. Based on the variable chart, the storage difference of the storage thermal insulation pipe in the current environment is calculated, and based on the storage difference, a quality system corresponding to the storage thermal insulation pipe is constructed;
[0008] Capturing images of the storage insulation pipes to obtain insulation pipe images, inputting the insulation pipe images into a pre-trained deep learning damage analysis model, and performing image enhancement processing on the insulation pipe images using the image enhancement layer in the deep learning damage analysis model to obtain target insulation pipe images;
[0009] Using the texture convolution layer in the deep learning loss analysis model to extract texture features of the target insulation pipe image, and obtaining the texture features of the insulation pipe; using the YUV convolution layer in the deep learning loss analysis model to extract color features of the target insulation pipe image, and obtaining the color features of the insulation pipe; using the contour extraction layer in the deep learning loss analysis model to extract contour features of the target insulation pipe image, and obtaining the contour features of the insulation pipe;
[0010] In combination with the texture features of the insulation pipe, the color features of the insulation pipe and the contour features of the insulation pipe, the residual analysis layer in the deep learning loss analysis model is used to calculate the loss coefficient corresponding to the storage insulation pipe;
[0011] In combination with the quality system, the depreciation coefficient and the sampling conditions, a sampling strategy corresponding to the insulation pipe is formulated, and according to the sampling strategy, the storage insulation pipes are sampled to obtain the sampling results.
[0012] Optionally, the environmental parameter data is analyzed and processed in combination with the sampling conditions to obtain parameter variables and variable observation value data, including:
[0013] Perform semantic analysis on the sampling conditions to obtain conditional semantics;
[0014] Analyzing the sampling target corresponding to the sampling condition according to the conditional semantics;
[0015] Performing data cleaning processing on the environmental parameter data to obtain cleaned parameter data;
[0016] Extracting the initial data variables corresponding to the cleaning parameter data, and calculating the correlation coefficient between the sampling target and the initial data variables;
[0017] According to the correlation coefficient, the initial parameter variable is filtered to obtain a parameter variable;
[0018] According to the parameter variables, the variable observation value data in the target parameter data is extracted.
[0019] Optionally, the calculating the correlation coefficient between the sampling target and the initial data variable includes:
[0020] The correlation coefficient between the sampling target and the initial data variable can be calculated by the following formula:
[0021]
[0022] Among them, Q represents the correlation coefficient between the sampling target and the initial data variable, R a Represents the vector corresponding to the ath target in the sampled target, R’ represents the target vector mean of the sampling target, a represents the target sequence number of the sampling target, T b Represents the vector corresponding to the bth variable in the initial data variable, T ’ represents the variable vector mean of the initial data variable, and b represents the variable sequence number of the initial data variable.
[0023] Optionally, calculating the storage difference of the storage insulation pipe in the current environment according to the variable chart includes:
[0024] Query the standard observation value corresponding to the storage insulation pipe;
[0025] The standard observation value is annotated in the variable chart to obtain a target variable chart;
[0026] According to the target variable chart, the standard deviation value corresponding to each variable in the environmental variables is calculated, and the variable weight corresponding to the environmental variables is assigned;
[0027] The storage difference of the warehouse insulation pipe in the current environment is calculated by combining the variable weight and the standard deviation value.
[0028] Optionally, performing image enhancement processing on the thermal insulation pipe image using the image enhancement layer in the deep learning damage analysis model to obtain a target thermal insulation pipe image includes:
[0029] De-noising the thermal insulation pipe image using the filter function in the image enhancement layer to obtain a de-noised thermal insulation pipe image;
[0030] Using the background removal algorithm in the image enhancement layer to perform background removal processing on the denoised insulation pipe image to obtain an insulation pipe foreground image;
[0031] Performing brightness optimization processing on the insulation pipe foreground image using the activation function in the image enhancement layer to obtain an optimized insulation pipe image;
[0032] The optimized thermal insulation pipe image is sharpened and enhanced using the sharpening algorithm in the image enhancement layer to obtain a target thermal insulation pipe image.
[0033] Optionally, the step of extracting texture features of the target thermal insulation pipe image using the texture convolution layer in the deep learning loss analysis model to obtain the thermal insulation pipe texture features includes:
[0034] Using the grayscale conversion layer in the texture convolution layer to perform grayscale conversion processing on the target thermal insulation pipe image to obtain a grayscale thermal insulation pipe image;
[0035] Using the texture filter in the texture convolution layer to perform texture convolution processing on the grayscale thermal insulation pipe image to obtain a texture feature map;
[0036] Using the normalization layer in the texture convolution layer to normalize the texture feature map, to obtain a target texture feature map;
[0037] Calculating the feature importance corresponding to the target texture feature map using the feature evaluation function in the texture convolution layer;
[0038] According to the feature importance, filtering out a key texture feature map from the target texture feature map;
[0039] The key texture feature map is pooled using the pooling layer in the texture convolution layer to obtain the texture features of the insulation pipe.
[0040] Optionally, the calculating the feature importance corresponding to the target texture feature map by using the feature evaluation function in the texture convolution layer includes:
[0041] The feature evaluation function includes:
[0042] B=H(E d )-∑(F d / n)*H(E d )
[0043] Among them, B represents the feature importance corresponding to the target texture feature map, H(E d ) represents the feature entropy value corresponding to the dth feature map in the target texture feature map, F d It represents the frequency of the dth feature map in the target texture feature map, and n represents the total number of target texture feature maps.
[0044] Optionally, the combining of the texture feature of the insulation pipe, the color feature of the insulation pipe and the contour feature of the insulation pipe and the use of the residual analysis layer in the deep learning loss analysis model to calculate the loss coefficient corresponding to the storage insulation pipe includes:
[0045] Performing feature coding processing on the texture feature of the thermal insulation pipe, the color feature of the thermal insulation pipe, and the contour feature of the thermal insulation pipe to obtain a first coding feature;
[0046] Obtain a standard insulation pipe image corresponding to the storage insulation pipe, and construct a feature extraction label corresponding to the standard insulation pipe image by combining the insulation pipe texture feature, the insulation pipe color feature and the insulation pipe contour feature;
[0047] In combination with the feature extraction label, extract the multi-dimensional features corresponding to the standard insulation pipe image;
[0048] Performing feature encoding processing on the multi-dimensional feature to obtain a second encoded feature;
[0049] In combination with the first coding feature and the second coding feature, the depreciation coefficient corresponding to the warehouse insulation pipe is calculated using the residual function in the residual analysis layer.
[0050] Optionally, the combining of the first coding feature and the second coding feature and using the residual function in the residual analysis layer to calculate the depreciation coefficient corresponding to the storage thermal insulation pipe includes:
[0051] The residual function includes:
[0052]
[0053] Among them, M represents the depreciation coefficient corresponding to the warehouse insulation pipe, G1 represents the vector value corresponding to the first coding feature, and G2 represents the vector value corresponding to the second coding feature. represents the feature similarity between the first encoding feature and the second encoding feature, and cossim() represents the cosine similarity function.
[0054] A storage insulation pipe sampling system based on deep learning, characterized in that the system includes:
[0055] A data acquisition module is used to obtain the storage insulation pipes to be sampled and the sampling conditions, and to collect the environmental parameter data corresponding to the storage insulation pipes;
[0056] A quality analysis module is used to analyze and process the environmental parameter data in combination with the sampling conditions to obtain environmental variables and variable observation data, and to construct a variable chart corresponding to the environmental variables based on the variable observation data, and to calculate the storage difference of the storage thermal insulation pipe in the current environment based on the variable chart, and to construct a quality system corresponding to the storage thermal insulation pipe based on the storage difference;
[0057] An image processing module is used to collect images of the storage insulation pipes to obtain insulation pipe images, input the insulation pipe images into a pre-trained deep learning loss analysis model, and perform image enhancement processing on the insulation pipe images using the image enhancement layer in the deep learning loss analysis model to obtain target insulation pipe images;
[0058] A feature extraction module is used to extract texture features of the target thermal insulation pipe image using the texture convolution layer in the deep learning loss analysis model to obtain the texture features of the thermal insulation pipe, to extract color features of the target thermal insulation pipe image using the YUV convolution layer in the deep learning loss analysis model to obtain the color features of the thermal insulation pipe, and to extract contour features of the target thermal insulation pipe image using the contour extraction layer in the deep learning loss analysis model to obtain the contour features of the thermal insulation pipe;
[0059] A loss coefficient calculation module, which is used to calculate the loss coefficient corresponding to the storage insulation pipe by combining the texture features of the insulation pipe, the color features of the insulation pipe and the contour features of the insulation pipe, and using the residual analysis layer in the deep learning loss analysis model;
[0060] The sampling strategy formulation module is used to formulate a sampling strategy corresponding to the insulation pipe in combination with the quality system, the loss coefficient and the sampling conditions, and sample the storage insulation pipe according to the sampling strategy to obtain the sampling results.
[0061] The present invention collects the environmental parameter data corresponding to the storage insulated pipe, and then obtains the specific data of the storage environment corresponding to the storage insulated pipe, so as to facilitate the subsequent storage sensitivity analysis of the storage insulated pipe. The present invention analyzes and processes the environmental parameter data in combination with the sampling conditions, and can evaluate and monitor the environmental conditions, and then obtain detailed environmental information in the environmental parameter data, providing a basis for the construction of subsequent variable charts. The present invention uses the image enhancement layer in the deep learning loss analysis model to perform image enhancement processing on the insulation pipe image, which can improve the image quality of the insulation pipe image and improve the accuracy of subsequent image feature extraction. The present invention uses the texture convolution layer in the deep learning loss analysis model to extract texture features from the target insulation pipe image, and can obtain the representation of the texture structure of different areas in the target insulation pipe image, which provides a basis for the subsequent loss analysis of the storage insulated pipe. The present invention formulates a sampling strategy corresponding to the insulation pipe by combining the quality system, the loss coefficient and the sampling conditions, thereby improving the sampling accuracy of the storage insulated pipe. Therefore, the embodiment of the present invention provides a method and system for sampling insulated pipes in storage based on deep learning, which can improve the sampling accuracy of insulated pipes in storage. BRIEF DESCRIPTION OF THE DRAWINGS
[0062] Figure 1 A schematic diagram of a flow chart of a method for sampling insulation pipes for storage based on deep learning provided in one embodiment of the present invention;
[0063] Figure 2A functional module diagram of an insulated pipe sampling system for warehousing based on deep learning provided in one embodiment of the present invention.
[0064] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0065] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0066] The embodiment of the present application provides a method for sampling insulated pipes for warehousing based on deep learning. In the embodiment of the present application, the execution subject of the method for sampling insulated pipes for warehousing based on deep learning includes but is not limited to at least one of the electronic devices such as the server and the terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for sampling insulated pipes for warehousing based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.
[0067] Reference Figure 1 FIG. 1 is a flow chart of a method for sampling insulated pipes for warehousing based on deep learning according to an embodiment of the present invention. In this embodiment, the method for sampling insulated pipes for warehousing based on deep learning includes steps S1-S6.
[0068] S1. Obtain the storage insulation pipes to be sampled and the sampling conditions, and collect the environmental parameter data corresponding to the storage insulation pipes.
[0069] The present invention collects the environmental parameter data corresponding to the storage insulated pipe, and then obtains the specific data of the storage environment corresponding to the storage insulated pipe, so as to facilitate the subsequent storage sensitivity analysis of the storage insulated pipe, wherein the storage insulated pipe is a pipe material used for insulation and heat insulation stored in a warehouse, usually used in the piping system in a facility, and it uses a special insulation material to cover the outside of the pipe to reduce heat transfer and loss, the sampling condition is the sampling requirement corresponding to the storage insulated pipe, and the environmental parameter data is the relevant data in the storage environment corresponding to the storage insulated pipe, such as humidity or temperature data, optionally, the collection of the environmental parameter data corresponding to the storage insulated pipe can be achieved by various environmental parameter sensors, such as temperature sensors, humidity sensors, pressure sensors, etc.
[0070] S2. In combination with the sampling conditions, the environmental parameter data is analyzed and processed to obtain parameter variables and variable observation data; based on the variable observation data, a variable chart corresponding to the parameter variables is constructed; based on the variable chart, the storage difference of the warehouse insulation pipe in the current environment is calculated; based on the storage difference, a quality system corresponding to the warehouse insulation pipe is constructed.
[0071] The present invention analyzes and processes the environmental parameter data in combination with the sampling conditions, thereby evaluating and monitoring the environmental conditions, and further obtaining detailed environmental information in the environmental parameter data, providing a basis for the construction of subsequent variable charts, wherein the parameter variable is the environmental type in the environmental parameter data, such as temperature, humidity or pressure, and the variable observation value data is the numerical data corresponding to the parameter variable.
[0072] As an embodiment of the present invention, the environmental parameter data is analyzed and processed in combination with the sampling conditions to obtain parameter variables and variable observation value data, including: performing semantic analysis on the sampling conditions to obtain conditional semantics, analyzing the sampling targets corresponding to the sampling conditions according to the conditional semantics, performing data cleaning processing on the environmental parameter data to obtain cleaned parameter data, extracting initial data variables corresponding to the cleaned parameter data, calculating the correlation coefficient between the sampling target and the initial data variables, filtering the initial parameter variables according to the correlation coefficient to obtain parameter variables, and extracting variable observation value data in the target parameter data according to the parameter variables.
[0073] Among them, the conditional semantics are the meaning and explanation corresponding to the sampling conditions, the sampling target is the expectation corresponding to the sampling conditions, the cleaning parameter data is the data obtained after removing missing data or duplicate data in the environmental parameter data, the initial data variables are the variables corresponding to all data in the cleaning parameter data, the association coefficient represents the degree of association between the sampling target and the initial data variables, and the parameter variable is a variable in the initial data variables that has a relationship with the sampling target.
[0074] Optionally, semantic analysis of the sampling condition can be achieved through a semantic analysis method, the sampling target corresponding to the sampling condition can be determined according to the semantic characters in the condition semantics, data cleaning processing of the environmental parameter data can be achieved through a cleaning tool, the cleaning tool is compiled by a scripting language, and extracting initial data variables corresponding to the cleaned parameter data can be achieved through a variable extraction tool, the variable extraction tool is compiled by JAVA language, filtering processing of the initial parameter variables can be achieved through a filtering function, such as a filter function, and extracting variable observation value data in the target parameter data can be achieved through an extraction function, and the extraction function is compiled by a programming language.
[0075] Optionally, as an optional embodiment of the present invention, the calculating the correlation coefficient between the sampling target and the initial data variable includes:
[0076] The correlation coefficient between the sampling target and the initial data variable can be calculated by the following formula:
[0077]
[0078] Among them, Q represents the correlation coefficient between the sampling target and the initial data variable, R a Represents the vector corresponding to the ath target in the sampled target, R ’ represents the target vector mean of the sampling target, a represents the target sequence number of the sampling target, T b Represents the vector corresponding to the bth variable in the initial data variable, T ’ represents the variable vector mean of the initial data variable, and b represents the variable sequence number of the initial data variable.
[0079] The present invention constructs a variable chart corresponding to the environmental variables based on the variable observation data, so that the data trend of the environmental variables can be intuitively understood through the variable chart, and the storage difference of the storage insulation pipe in the current environment can be calculated. The storage difference can be used to understand the degree of storage difference between the current environment and the standard environment of the storage insulation pipe, thereby facilitating the subsequent quality analysis of the storage insulation pipe, wherein the variable chart is a visualization chart corresponding to the environmental variables, such as a line chart or a scatter plot, and the storage difference indicates the degree of difference in storage effect of the storage insulation pipe between the current environment and the standard storage environment. Optionally, constructing the variable chart corresponding to the environmental variables can be achieved through a drawing tool, such as a Visio drawing tool.
[0080] As an embodiment of the present invention, the storage difference of the storage insulated pipe in the current environment is calculated based on the variable chart, including: querying the standard observation value corresponding to the storage insulated pipe, annotating the standard observation value in the variable chart to obtain a target variable chart, calculating the standard deviation value corresponding to each variable in the environmental variables according to the target variable chart, and assigning variable weights corresponding to the environmental variables, and calculating the storage difference of the storage insulated pipe in the current environment by combining the variable weights and the standard deviation values.
[0081] Among them, the standard observation value is the standard numerical value corresponding to the variable of the standard storage environment corresponding to the warehouse insulation pipe, the target variable chart is a chart obtained by marking the standard observation value in the corresponding chart, the standard deviation value represents the difference between each variable in the environmental variable and the standard variable, and the variable weight represents the importance corresponding to the environmental variable.
[0082] Optionally, the standard observation value corresponding to the warehouse insulation pipe can be queried from the Internet in a human-computer interaction manner, and the fitting standard curve is obtained by fitting the points corresponding to the standard observation values in the target variable chart, and the fitting observation curve is obtained by fitting the points corresponding to the variable observation values in the target variable chart, and the integrals corresponding to the fitted standard curve and the fitted observation curve are calculated respectively to obtain a first integral and a second integral, and the integral difference between the first integral and the second integral is calculated to obtain the standard deviation value corresponding to each variable in the environmental variables, and the quality evaluation index corresponding to the warehouse insulation pipe is queried, and the vector similarity between the environmental variables and the quality evaluation index is calculated. According to the vector similarity, the variable weight corresponding to the environmental variable is assigned, and the storage deviation value can be obtained by calculating the variable weight and the standard deviation value using the weighted sum method.
[0083] The present invention constructs a quality system corresponding to the storage insulated pipe according to the storage difference, and then different quality levels of the heating of the storage insulated pipe can be obtained, which provides a basis for the formulation of subsequent sampling strategies, wherein the quality system is the quality level corresponding to the storage insulated pipe. Optionally, the storage difference is sorted in descending order to obtain a difference sequence. According to the difference sequence, the quality system of the storage insulated pipe is constructed. For example, the greater the storage difference, the greater the impact on the storage insulated pipe corresponding to the area, which leads to a greater decrease in the quality of the storage insulated pipe. The quality level of the storage insulated pipe is set in this way, and then the quality system corresponding to the storage insulated pipe is obtained.
[0084] S3. Capture images of the warehouse insulated pipes to obtain an insulated pipe image, input the insulated pipe image into a pre-trained deep learning loss analysis model, and use the image enhancement layer in the deep learning loss analysis model to perform image enhancement processing on the insulated pipe image to obtain a target insulated pipe image.
[0085] The present invention performs image enhancement processing on the insulated pipe image by utilizing the image enhancement layer in the deep learning loss analysis model, thereby improving the image quality of the insulated pipe image and improving the accuracy of subsequent image feature extraction, wherein the insulated pipe image is an image corresponding to the stored insulated pipe, and the deep learning loss analysis model is a model that uses deep learning technology to evaluate the loss condition of the stored insulated pipe. The target insulated pipe image is an image obtained after the quality of the insulated pipe image is enhanced. Optionally, image acquisition of the stored insulated pipe can be achieved by an image acquisition device, such as a camera device.
[0086] As an embodiment of the present invention, the image enhancement layer in the deep learning loss analysis model is used to perform image enhancement processing on the insulation pipe image to obtain a target insulation pipe image, including: using the filtering function in the image enhancement layer to perform denoising processing on the insulation pipe image to obtain a denoised insulation pipe image, using the background removal algorithm in the image enhancement layer to perform background removal processing on the denoised insulation pipe image to obtain an insulation pipe foreground image, using the activation function in the image enhancement layer to perform brightness optimization processing on the insulation pipe foreground image to obtain an optimized insulation pipe image, and using the sharpening algorithm in the image enhancement layer to perform sharpening and enhancement processing on the optimized insulation pipe image to obtain a target insulation pipe image.
[0087] Among them, the filtering function is a function used to remove noise interference in the insulation pipe image, such as a Gaussian filter; the background removal algorithm is an algorithm used to remove the background in the denoised insulation pipe image, such as a threshold segmentation method; the optimized insulation pipe image is an image obtained after the image brightness of the insulation pipe foreground image is enhanced; the sharpening algorithm is an algorithm used to enhance the image detail information in the optimized insulation pipe image, such as a Plasma operator; optionally, the insulation pipe foreground image can be nonlinearly transformed by the activation function to adjust the brightness and contrast of the insulation pipe foreground image, thereby achieving the purpose of brightness optimization; the activation function includes a Sigmoid function.
[0088] S4. Use the texture convolution layer in the deep learning loss analysis model to extract texture features of the target insulation pipe image to obtain the texture features of the insulation pipe. Use the YUV convolution layer in the deep learning loss analysis model to extract color features of the target insulation pipe image to obtain the color features of the insulation pipe. Use the contour extraction layer in the deep learning loss analysis model to extract contour features of the target insulation pipe image to obtain the contour features of the insulation pipe.
[0089] The present invention utilizes the texture convolution layer in the deep learning loss analysis model to extract texture features of the target insulated pipe image, thereby obtaining a representation of the texture structure of different areas in the target insulated pipe image, providing a basis for subsequent loss analysis of the warehouse insulated pipe, wherein the texture convolution layer is a neural network layer for extracting the texture representation of the target insulated pipe image, and the insulated pipe texture feature is a representation of the texture structure in the target insulated pipe image.
[0090] As an embodiment of the present invention, the texture convolution layer in the deep learning loss analysis model is used to extract texture features of the target insulation pipe image to obtain the insulation pipe texture features, including: using the grayscale conversion layer in the texture convolution layer to perform grayscale conversion processing on the target insulation pipe image to obtain a grayscale insulation pipe image, using the texture filter in the texture convolution layer to perform texture convolution processing on the grayscale insulation pipe image to obtain a texture feature map, using the normalization layer in the texture convolution layer to normalize the texture feature map to obtain a target texture feature map, using the feature evaluation function in the texture convolution layer to calculate the feature importance corresponding to the target texture feature map, according to the feature importance, screening out key texture feature maps from the target texture feature map, using the pooling layer in the texture convolution layer to pool the key texture feature maps to obtain the insulation pipe texture features.
[0091] Among them, the grayscale conversion layer is a neural layer used to convert the target insulation tube image into a single color image, which is composed of the cv2.cvtColor() function. The grayscale insulation tube image is an image of the target insulation tube image expressed in a single color. The texture filter is a filter used to extract texture features of the grayscale insulation tube image, such as a Gabor filter. The texture feature map is an image representing image texture information in the grayscale insulation tube image. The target texture feature map is a feature map after the texture feature map is converted into a unified scale standard. The feature importance represents a measure of the importance of the target texture feature map.
[0092] Optionally, normalization of the texture feature map can be achieved by a normalization function in the normalization layer, such as a Z-score normalization function, and the feature importance is compared with a standard entropy value. If the feature importance is greater than the standard entropy value, the target texture feature map corresponding to the feature importance is screened out to obtain a key texture feature map. The key texture feature map can be pooled by a pooling function in the pooling layer, and the pooling function includes a maximum pooling function.
[0093] Optionally, as an optional embodiment of the present invention, the calculating the feature importance corresponding to the target texture feature map by using the feature evaluation function in the texture convolution layer includes:
[0094] The feature evaluation function includes:
[0095] B=H(E d )-Σ(F d / n)*H(E d )
[0096] Among them, B represents the feature importance corresponding to the target texture feature map, H(E d ) represents the feature entropy value corresponding to the dth feature map in the target texture feature map, F d It represents the frequency of the dth feature map in the target texture feature map, and n represents the total number of target texture feature maps.
[0097] The present invention uses the YUV convolution layer in the deep learning loss analysis model to extract color features of the target insulated pipe image, so as to obtain color detail information in the target insulated pipe image, and uses the contour extraction layer in the deep learning loss analysis model to extract contour features of the target insulated pipe image, so as to obtain shape features in the target insulated pipe image, thereby improving the accuracy of subsequent calculation of the loss coefficient corresponding to the warehouse insulated pipe, wherein the YUV convolution layer is a convolution layer for extracting features of the color of the target insulated pipe image, and the YUV convolution layer can be used to extract the features of the color of the target insulated pipe image. The brightness (Y) and chromaticity (U, V) of the target insulated pipe image are separated, the insulated pipe color feature is the color representation corresponding to the target insulated pipe image, the contour extraction layer is a neural network for extracting the shape representation of the insulated pipe in the target insulated pipe image, and the insulated pipe contour feature is the shape representation of the insulated pipe in the target insulated pipe image. Optionally, the color feature of the target insulated pipe image can be extracted by the YUV convolution kernel in the YUV convolution layer, and the contour feature of the target insulated pipe image can be extracted by the Canny edge detection algorithm in the contour extraction layer.
[0098] S5. Combining the texture features, color features and contour features of the insulation pipe, the residual analysis layer in the deep learning loss analysis model is used to calculate the loss coefficient corresponding to the storage insulation pipe.
[0099] The present invention combines the texture features of the insulation pipe, the color features of the insulation pipe and the contour features of the insulation pipe, and uses the residual analysis layer in the deep learning depreciation analysis model to calculate the depreciation coefficient corresponding to the storage insulation pipe. The depreciation degree of the storage insulation pipe can be understood, thereby judging the degree of appearance wear of the storage insulation pipe.
[0100] As an embodiment of the present invention, the insulation pipe texture features, the insulation pipe color features and the insulation pipe contour features are combined, and the residual analysis layer in the deep learning loss analysis model is used to calculate the loss coefficient corresponding to the warehouse insulation pipe, including: performing feature encoding processing on the insulation pipe texture features, the insulation pipe color features and the insulation pipe contour features to obtain a first encoding feature, obtaining a standard insulation pipe image corresponding to the warehouse insulation pipe, combining the insulation pipe texture features, the insulation pipe color features and the insulation pipe contour features to construct a feature extraction label corresponding to the standard insulation pipe image, combining the feature extraction label to extract multi-dimensional features corresponding to the standard insulation pipe image, performing feature encoding processing on the multi-dimensional features to obtain a second encoding feature, combining the first encoding feature and the second encoding feature, and using the residual function in the residual analysis layer to calculate the loss coefficient corresponding to the warehouse insulation pipe.
[0101] Among them, the first coding feature is the feature vector obtained by converting the insulation pipe texture feature, the insulation pipe color feature and the insulation pipe contour feature into corresponding vectors, and fusing the corresponding vectors; the standard insulation pipe image is the image of the reference standard corresponding to the warehouse insulation pipe; the feature extraction label is the feature extraction type of the standard insulation pipe image; the multi-dimensional feature is a collection of texture, color and contour features of the standard insulation pipe image; the second coding feature is the fused feature vector corresponding to the multi-dimensional feature; and the residual function is a function used to calculate the depreciation coefficient of the warehouse insulation pipe.
[0102] Optionally, the texture features, color features and contour features of the insulation pipe can be vectorized using the word2vec algorithm to obtain corresponding feature vectors, and the feature vectors can be encoded and embedded to obtain first coding features. The standard insulation pipe image can be obtained through the technical documents provided by the manufacturer of the warehouse insulation pipe, and the feature extraction labels corresponding to the standard insulation pipe image can be constructed through a label generator, which is compiled by a scripting language.
[0103] Optionally, as an embodiment of the present invention, the combining of the first coding feature and the second coding feature and the use of the residual function in the residual analysis layer to calculate the depreciation coefficient corresponding to the storage thermal insulation pipe include:
[0104] The residual function includes:
[0105]
[0106] Among them, M represents the depreciation coefficient corresponding to the warehouse insulation pipe, G1 represents the vector value corresponding to the first coding feature, and G2 represents the vector value corresponding to the second coding feature. represents the feature similarity between the first encoding feature and the second encoding feature, and cossim() represents the cosine similarity function.
[0107] S6. Based on the quality system, the depreciation coefficient and the sampling conditions, a sampling strategy corresponding to the insulation pipes is formulated. According to the sampling strategy, the storage insulation pipes are sampled to obtain sampling results.
[0108] The present invention formulates a sampling strategy corresponding to the insulated pipes by combining the quality system, the loss coefficient and the sampling conditions, thereby improving the sampling accuracy of the stored insulated pipes, wherein the sampling strategy is a detailed sampling method corresponding to the stored insulated pipes. Optionally, the sampling quantity of the stored insulated pipes is determined according to the sampling conditions, and the stored insulated pipes are layered according to the quality system to obtain layered insulated pipes. According to the loss coefficient, the sampling ratio corresponding to the layered insulated pipes is set, and the sampling strategy for the stored insulated pipes is formulated in combination with the sampling quantity and the sampling ratio.
[0109] The present invention collects the environmental parameter data corresponding to the storage insulated pipe, and then obtains the specific data of the storage environment corresponding to the storage insulated pipe, so as to facilitate the subsequent storage sensitivity analysis of the storage insulated pipe. The present invention analyzes and processes the environmental parameter data in combination with the sampling conditions, and can evaluate and monitor the environmental conditions, and then obtain detailed environmental information in the environmental parameter data, providing a basis for the construction of subsequent variable charts. The present invention uses the image enhancement layer in the deep learning loss analysis model to perform image enhancement processing on the insulation pipe image, which can improve the image quality of the insulation pipe image and improve the accuracy of subsequent image feature extraction. The present invention uses the texture convolution layer in the deep learning loss analysis model to extract texture features from the target insulation pipe image, and can obtain the representation of the texture structure of different areas in the target insulation pipe image, which provides a basis for the subsequent loss analysis of the storage insulated pipe. The present invention formulates a sampling strategy corresponding to the insulation pipe by combining the quality system, the loss coefficient and the sampling conditions, thereby improving the sampling accuracy of the storage insulated pipe. Therefore, the insulated pipe sampling method for storage based on deep learning provided in an embodiment of the present invention can improve the sampling accuracy of the insulated pipes in storage.
[0110] like Figure 2 As shown, it is a functional module diagram of an insulated pipe sampling system for warehousing based on deep learning provided by one embodiment of the present invention.
[0111] The insulated pipe sampling system 100 for warehousing based on deep learning described in the present invention can be installed in an electronic device. According to the functions implemented, the insulated pipe sampling system 100 for warehousing based on deep learning can include a data acquisition module 101, a quality analysis module 102, an image processing module 103, a feature extraction module 104, a loss coefficient calculation module 105 and a sampling strategy formulation module 106. The module described in the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.
[0112] In this embodiment, the functions of each module / unit are as follows:
[0113] The data acquisition module 101 is used to obtain the storage insulation pipes to be sampled and the sampling conditions, and to collect the environmental parameter data corresponding to the storage insulation pipes;
[0114] The quality analysis module 102 is used to analyze and process the environmental parameter data in combination with the sampling conditions to obtain environmental variables and variable observation data, and to construct a variable chart corresponding to the environmental variables based on the variable observation data, and to calculate the storage difference of the storage thermal insulation pipe in the current environment based on the variable chart, and to construct a quality system corresponding to the storage thermal insulation pipe based on the storage difference;
[0115] The image processing module 103 is used to collect images of the storage insulation pipes to obtain insulation pipe images, input the insulation pipe images into a pre-trained deep learning damage analysis model, and perform image enhancement processing on the insulation pipe images using the image enhancement layer in the deep learning damage analysis model to obtain target insulation pipe images;
[0116] The feature extraction module 104 is used to extract texture features of the target insulation pipe image using the texture convolution layer in the deep learning loss analysis model to obtain the insulation pipe texture features, to extract color features of the target insulation pipe image using the YUV convolution layer in the deep learning loss analysis model to obtain the insulation pipe color features, and to extract contour features of the target insulation pipe image using the contour extraction layer in the deep learning loss analysis model to obtain the insulation pipe contour features;
[0117] The depreciation coefficient calculation module 105 is used to calculate the depreciation coefficient corresponding to the storage insulation pipe by using the residual analysis layer in the deep learning depreciation analysis model in combination with the texture feature of the insulation pipe, the color feature of the insulation pipe and the contour feature of the insulation pipe;
[0118] The sampling strategy formulation module 106 is used to formulate a sampling strategy corresponding to the insulation pipe in combination with the quality system, the loss coefficient and the sampling conditions, and sample the storage insulation pipes according to the sampling strategy to obtain sampling results.
[0119] In detail, each module described in the insulated pipe sampling system 100 for implementing storage based on deep learning described in the embodiment of the present application is used in the same manner as described above. Figure 1 The technical means described in the article are the same as the deep learning-based sampling method for storage insulation pipes, and can produce the same technical effects, so I will not go into details here.
[0120] In the several embodiments provided by the present invention, it should be understood that the provided methods and systems can be implemented in other ways. For example, the method embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.
[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A method for sampling insulation pipes for storage based on deep learning, characterized in that: The method comprises: Obtain the storage insulation pipes to be sampled and the sampling conditions, and collect the environmental parameter data corresponding to the storage insulation pipes; In combination with the sampling conditions, the environmental parameter data is analyzed and processed to obtain environmental variables and variable observation data, and a variable chart corresponding to the environmental variables is constructed based on the variable observation data. Based on the variable chart, the storage difference of the storage thermal insulation pipe in the current environment is calculated, and based on the storage difference, a quality system corresponding to the storage thermal insulation pipe is constructed; Capturing images of the storage insulation pipes to obtain insulation pipe images, inputting the insulation pipe images into a pre-trained deep learning damage analysis model, and performing image enhancement processing on the insulation pipe images using the image enhancement layer in the deep learning damage analysis model to obtain target insulation pipe images; Using the texture convolution layer in the deep learning loss analysis model to extract texture features of the target insulation pipe image, and obtaining the texture features of the insulation pipe; using the YUV convolution layer in the deep learning loss analysis model to extract color features of the target insulation pipe image, and obtaining the color features of the insulation pipe; using the contour extraction layer in the deep learning loss analysis model to extract contour features of the target insulation pipe image, and obtaining the contour features of the insulation pipe; In combination with the texture features of the insulation pipe, the color features of the insulation pipe and the contour features of the insulation pipe, the residual analysis layer in the deep learning loss analysis model is used to calculate the loss coefficient corresponding to the storage insulation pipe; In combination with the quality system, the depreciation coefficient and the sampling conditions, a sampling strategy corresponding to the insulation pipe is formulated, and according to the sampling strategy, the storage insulation pipes are sampled to obtain the sampling results.
2. A method for sampling insulation pipes for warehousing based on deep learning as claimed in claim 1, characterized in that: The environmental parameter data is analyzed and processed in combination with the sampling conditions to obtain parameter variables and variable observation value data, including: Perform semantic analysis on the sampling conditions to obtain conditional semantics; Analyzing the sampling target corresponding to the sampling condition according to the conditional semantics; Performing data cleaning processing on the environmental parameter data to obtain cleaned parameter data; Extracting the initial data variables corresponding to the cleaning parameter data, and calculating the correlation coefficient between the sampling target and the initial data variables; According to the correlation coefficient, the initial parameter variable is filtered to obtain a parameter variable; According to the parameter variables, the variable observation value data in the target parameter data is extracted.
3. A method for sampling thermal insulation pipes for warehousing based on deep learning as claimed in claim 2, characterized in that: The calculating the correlation coefficient between the sampling target and the initial data variable comprises: The correlation coefficient between the sampling target and the initial data variable can be calculated by the following formula: Among them, Q represents the correlation coefficient between the sampling target and the initial data variable, R a Represents the vector corresponding to the ath target in the sampled target, R ’ represents the target vector mean of the sampling target, a represents the target sequence number of the sampling target, T b Represents the vector corresponding to the bth variable in the initial data variable, T ’ represents the variable vector mean of the initial data variable, and b represents the variable sequence number of the initial data variable.
4. The method for sampling thermal insulation pipes for warehousing based on deep learning according to claim 1, characterized in that: The step of calculating the storage difference of the storage insulation pipe in the current environment according to the variable chart includes: Query the standard observation value corresponding to the storage insulation pipe; The standard observation value is annotated in the variable chart to obtain a target variable chart; According to the target variable chart, the standard deviation value corresponding to each variable in the environmental variables is calculated, and the variable weight corresponding to the environmental variables is assigned; The storage difference of the warehouse insulation pipe in the current environment is calculated by combining the variable weight and the standard deviation value.
5. The method for sampling thermal insulation pipes for warehousing based on deep learning as claimed in claim 1, characterized in that: The method of performing image enhancement processing on the thermal insulation pipe image by using the image enhancement layer in the deep learning damage analysis model to obtain a target thermal insulation pipe image includes: De-noising the thermal insulation pipe image using the filter function in the image enhancement layer to obtain a de-noised thermal insulation pipe image; Using the background removal algorithm in the image enhancement layer to perform background removal processing on the denoised insulation pipe image to obtain an insulation pipe foreground image; Performing brightness optimization processing on the insulation pipe foreground image using the activation function in the image enhancement layer to obtain an optimized insulation pipe image; The optimized thermal insulation pipe image is sharpened and enhanced using the sharpening algorithm in the image enhancement layer to obtain a target thermal insulation pipe image.
6. The method for sampling thermal insulation pipes for warehousing based on deep learning as claimed in claim 1, characterized in that: The method of extracting texture features of the target insulation pipe image using the texture convolution layer in the deep learning loss analysis model to obtain the insulation pipe texture features includes: Using the grayscale conversion layer in the texture convolution layer to perform grayscale conversion processing on the target thermal insulation pipe image to obtain a grayscale thermal insulation pipe image; Using the texture filter in the texture convolution layer to perform texture convolution processing on the grayscale thermal insulation pipe image to obtain a texture feature map; Using the normalization layer in the texture convolution layer to normalize the texture feature map, to obtain a target texture feature map; Calculating the feature importance corresponding to the target texture feature map using the feature evaluation function in the texture convolution layer; According to the feature importance, a key texture feature map is screened out from the target texture feature map; The key texture feature map is pooled using the pooling layer in the texture convolution layer to obtain the texture features of the insulation pipe.
7. A method for sampling thermal insulation pipes for storage based on deep learning as claimed in claim 6, characterized in that: The calculating the feature importance corresponding to the target texture feature map by using the feature evaluation function in the texture convolution layer includes: The feature evaluation function includes: B=H(E d )-Σ(F d / n)*H(E d ) Among them, B represents the feature importance corresponding to the target texture feature map, H(E d ) represents the feature entropy value corresponding to the dth feature map in the target texture feature map, F d It represents the frequency of the dth feature map in the target texture feature map, and n represents the total number of target texture feature maps.
8. The method for sampling thermal insulation pipes for warehousing based on deep learning as claimed in claim 1, characterized in that: The method combines the texture feature of the insulation pipe, the color feature of the insulation pipe and the contour feature of the insulation pipe, and uses the residual analysis layer in the deep learning loss analysis model to calculate the loss coefficient corresponding to the storage insulation pipe, including: Performing feature coding processing on the texture feature of the thermal insulation pipe, the color feature of the thermal insulation pipe, and the contour feature of the thermal insulation pipe to obtain a first coding feature; Obtain a standard thermal insulation pipe image corresponding to the storage thermal insulation pipe, and construct a feature extraction label corresponding to the standard thermal insulation pipe image by combining the thermal insulation pipe texture feature, the thermal insulation pipe color feature, and the thermal insulation pipe contour feature; In combination with the feature extraction label, extract the multi-dimensional features corresponding to the standard insulation pipe image; Performing feature encoding processing on the multi-dimensional feature to obtain a second encoded feature; In combination with the first coding feature and the second coding feature, the depreciation coefficient corresponding to the warehouse insulation pipe is calculated using the residual function in the residual analysis layer.
9. A method for sampling insulation pipes for warehousing based on deep learning as claimed in claim 8, characterized in that: The combining of the first coding feature and the second coding feature and using the residual function in the residual analysis layer to calculate the depreciation coefficient corresponding to the storage thermal insulation pipe includes: The residual function includes: Among them, M represents the depreciation coefficient corresponding to the warehouse insulation pipe, G1 represents the vector value corresponding to the first coding feature, and G2 represents the vector value corresponding to the second coding feature. represents the feature similarity between the first encoding feature and the second encoding feature, and cossim() represents the cosine similarity function.
10. A deep learning based storage insulation pipe sampling system, characterized in that: The system comprises: A data acquisition module is used to obtain the storage insulation pipes to be sampled and the sampling conditions, and to collect the environmental parameter data corresponding to the storage insulation pipes; A quality analysis module is used to analyze and process the environmental parameter data in combination with the sampling conditions to obtain environmental variables and variable observation data, and to construct a variable chart corresponding to the environmental variables based on the variable observation data, and to calculate the storage difference of the storage thermal insulation pipe in the current environment based on the variable chart, and to construct a quality system corresponding to the storage thermal insulation pipe based on the storage difference; An image processing module is used to collect images of the storage insulation pipes to obtain insulation pipe images, input the insulation pipe images into a pre-trained deep learning loss analysis model, and perform image enhancement processing on the insulation pipe images using the image enhancement layer in the deep learning loss analysis model to obtain target insulation pipe images; A feature extraction module is used to extract texture features of the target thermal insulation pipe image using the texture convolution layer in the deep learning loss analysis model to obtain the texture features of the thermal insulation pipe, to extract color features of the target thermal insulation pipe image using the YUV convolution layer in the deep learning loss analysis model to obtain the color features of the thermal insulation pipe, and to extract contour features of the target thermal insulation pipe image using the contour extraction layer in the deep learning loss analysis model to obtain the contour features of the thermal insulation pipe; A loss coefficient calculation module, which is used to calculate the loss coefficient corresponding to the storage insulation pipe by combining the texture features of the insulation pipe, the color features of the insulation pipe and the contour features of the insulation pipe, and using the residual analysis layer in the deep learning loss analysis model; The sampling strategy formulation module is used to formulate a sampling strategy corresponding to the insulation pipe in combination with the quality system, the loss coefficient and the sampling conditions, and sample the storage insulation pipe according to the sampling strategy to obtain the sampling results.
