WO2024103613A1 - 一种电子级六氟丁二烯的质检系统及其方法 - Google Patents

一种电子级六氟丁二烯的质检系统及其方法 Download PDF

Info

Publication number
WO2024103613A1
WO2024103613A1 PCT/CN2023/086973 CN2023086973W WO2024103613A1 WO 2024103613 A1 WO2024103613 A1 WO 2024103613A1 CN 2023086973 W CN2023086973 W CN 2023086973W WO 2024103613 A1 WO2024103613 A1 WO 2024103613A1
Authority
WO
WIPO (PCT)
Prior art keywords
graph
feature
attention
hexafluorobutadiene
grade
Prior art date
Application number
PCT/CN2023/086973
Other languages
English (en)
French (fr)
Inventor
练钢
邱桂祥
张国聪
罗昊卿
黄媛玲
Original Assignee
福建省杭氟电子材料有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 福建省杭氟电子材料有限公司 filed Critical 福建省杭氟电子材料有限公司
Publication of WO2024103613A1 publication Critical patent/WO2024103613A1/zh

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
    • G01N30/02—Column chromatography
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
    • G01N30/02—Column chromatography
    • G01N30/86—Signal analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the present application relates to the technical field of gas quality inspection, and more specifically, to a quality inspection system and method for electronic grade hexafluorobutadiene.
  • Electronic grade hexafluorobutadiene is a green and environmentally friendly dry etching gas with excellent etching performance. It is widely used in the production process of large-scale integrated circuits and high-speed and high-capacity storage chips. Among them, the moisture index in electronic grade hexafluorobutadiene has a serious impact on the optimization and improvement of integrated circuit and chip processes. Therefore, establishing a reliable analysis technology for moisture in electronic grade hexafluorobutadiene is the key to ensuring the quality of electronic grade hexafluorobutadiene products.
  • dew point method there are three methods for determining trace moisture in gas in the current national standard: dew point method, electrolysis method and cavity ring-down method.
  • the dew point method is simple to operate, but electronic grade hexafluorobutadiene will react with alumina and damage the instrument, so the dew point method cannot meet the requirements.
  • the electrolysis method and cavity ring-down method since the boiling point of electronic grade hexafluorobutadiene is 6°C, it is easy to liquefy and thus contaminate the hydrogen flame ionization detector, so the electrolysis method and cavity ring-down method are not applicable.
  • the embodiment of the present application provides a quality inspection system and method for electronic-grade hexafluorobutadiene, which uses an artificial intelligence algorithm based on deep learning to extract multi-scale implicit feature distribution information of different sizes in the gas chromatogram of electronic-grade hexafluorobutadiene, and uses an attention mechanism to further focus on the small-size water content detection in the electronic-grade hexafluorobutadiene and filter out useless interference feature information.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of electronic-grade hexafluorobutadiene.
  • a quality inspection system for electronic-grade hexafluorobutadiene comprising: a gas chromatogram acquisition module, used to obtain a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured; a feature detection module, used to pass the gas chromatogram through a first convolutional neural network model including a significant detector to obtain a gas chromatogram feature graph; a feature enhancement module, used to pass the gas chromatogram feature graph through a residual double attention mechanism model to obtain an enhanced gas chromatogram feature graph as a decoding feature graph; and a decoding module, used to pass the decoded feature graph through a decoder to obtain a decoding value, wherein the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured.
  • a quality inspection method for electronic-grade hexafluorobutadiene comprising: obtaining a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured; passing the gas chromatogram through a first convolutional neural network model including a significant detector to obtain a gas chromatogram characteristic graph; passing the gas chromatogram characteristic graph through a residual double attention mechanism model to obtain an enhanced gas chromatogram characteristic graph as a decoding characteristic graph; and passing the decoded characteristic graph through a decoder to obtain a decoding value, wherein the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured.
  • an electronic device comprising: a processor; and a memory, wherein computer program instructions are stored in the memory, and when the computer program instructions are executed by the processor, the processor executes the quality inspection method for electronic-grade hexafluorobutadiene as described above.
  • a computer-readable medium on which computer program instructions are stored.
  • the processor executes the quality inspection method for electronic-grade hexafluorobutadiene as described above.
  • the electronic-grade hexafluorobutadiene quality inspection system and method provided by the present application use an artificial intelligence algorithm based on deep learning to extract multi-scale implicit feature distribution information of different sizes in the gas chromatogram of the electronic-grade hexafluorobutadiene, and use an attention mechanism to further focus on the small-size water content detection in the electronic-grade hexafluorobutadiene and filter out useless interference feature information.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of the electronic-grade hexafluorobutadiene.
  • FIG1 illustrates an application scenario diagram of a quality inspection system and method for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG2 illustrates a block diagram of a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG3 illustrates a block diagram of a feature enhancement module in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG4 illustrates a block diagram of a spatial attention unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG5 illustrates a block diagram of a channel attention unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG6 illustrates a block diagram of a residual fusion unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG. 7 illustrates a flow chart of a quality inspection method for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG8 is a schematic diagram of a system architecture of a quality inspection method for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • FIG. 9 illustrates a flow chart of a method for quality inspection of electronic grade hexafluorobutadiene according to an embodiment of the present application, in which the gas chromatographic characteristic graph is subjected to a residual dual attention mechanism model to obtain an enhanced gas chromatographic characteristic graph as a decoding characteristic graph.
  • FIG. 10 illustrates a block diagram of an electronic device according to an embodiment of the present application.
  • electronic grade hexafluorobutadiene is a green and environmentally friendly dry etching gas with excellent etching performance, which is widely used in the production process of large-scale integrated circuits and high-speed and high-capacity storage chips.
  • the moisture index in electronic grade hexafluorobutadiene has a serious impact on the optimization and improvement of integrated circuit and chip processes. Therefore, establishing a reliable analysis technology for moisture in electronic grade hexafluorobutadiene is the key to ensuring the quality of electronic grade hexafluorobutadiene products.
  • dew point method there are three methods for determining trace moisture in gas in the current national standard: dew point method, electrolysis method and cavity ring-down method.
  • the dew point method is simple to operate, but electronic grade hexafluorobutadiene will react with alumina and damage the instrument, so the dew point method cannot meet the requirements.
  • the electrolysis method and cavity ring-down method since the boiling point of electronic grade hexafluorobutadiene is 6°C, it is easy to liquefy and thus contaminate the hydrogen flame ionization detector, so the electrolysis method and cavity ring-down method are not applicable. Therefore, an optimized quality inspection scheme for electronic grade hexafluorobutadiene is expected.
  • deep learning and neural networks have been widely used in computer vision, natural language processing, speech signal processing and other fields.
  • deep learning and neural networks have also shown a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields.
  • the chromatogram refers to the image of the detection signal of the separated components distributed over time
  • the water content of electronic-grade hexafluorobutadiene can be obtained from the gas chromatogram, but because water only accounts for a small part of electronic-grade hexafluorobutadiene, it is difficult to obtain an accurate water content measurement value through the gas chromatogram.
  • an artificial intelligence algorithm based on deep learning is used to extract multi-scale implicit feature distribution information of different sizes in the gas chromatogram of electronic-grade hexafluorobutadiene, and an attention mechanism is used to further focus on the small-size water content detection in the electronic-grade hexafluorobutadiene and filter out useless interference feature information.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of electronic-grade hexafluorobutadiene.
  • a gas chromatogram of electronic-grade hexafluorobutadiene to be measured is obtained.
  • a convolutional neural network model with excellent performance in implicit feature extraction of images is used to perform feature mining of the gas chromatogram of the electronic-grade hexafluorobutadiene to be measured, but considering that the convolutional neural network has poor resolution for small-scale features in images (for example, the amount of water in the electronic-grade hexafluorobutadiene to be measured in the embodiment of the present application).
  • the network structure of the standard convolutional neural network is adjusted, specifically, a first convolutional neural network model including a significant detector is used to encode the gas chromatogram to obtain a gas chromatogram characteristic graph.
  • each layer of the significant detector uses a combination of large and small convolutional kernels to enable each layer of the convolutional neural network model to pay more attention to small-sized object features.
  • each layer of the convolutional neural network uses a two-dimensional convolution kernel to perform a convolution encoding on the input data. Accordingly, in the technical solution of the present application, in order to improve the ability of each layer of the convolutional neural network to extract local significant features, each layer of the convolutional neural network is transformed to perform secondary convolution, and the size of the first convolution kernel is larger than the size of the second convolution kernel.
  • convolution through a larger convolution kernel has a larger receptive field, but the extracted image feature pattern is rough, and it is easy to ignore the details with resolution in the gas chromatogram, such as the implicit features of moisture; accordingly, using a smaller convolution kernel for secondary convolution can better model local information, and thus can focus on more detailed feature information.
  • the size of the first convolution kernel is 5 ⁇ 5
  • the size of the second convolution kernel is 3 ⁇ 3.
  • the gas chromatogram characteristic graph is further enhanced in feature data to obtain an enhanced gas chromatogram characteristic graph.
  • the network will obtain partial feature information, but will not automatically distinguish the detailed information between high and low frequencies and the differences between the features of each category.
  • the network's ability to selectively use features is limited.
  • the attention mechanism can select the focus position, it produces a more distinguishable feature representation, and the features after the attention module is added will produce adaptive changes as the network deepens. Therefore, in the technical solution of the present application, a channel attention mechanism and a spatial attention mechanism are introduced, and by introducing a residual structure, it is combined with the proposed dual attention network to construct a residual dual attention model.
  • This model combines spatial attention and channel attention in parallel, so that different types of effective information are captured in large quantities, which can effectively enhance the feature recognition learning ability.
  • the task processing system is more focused on finding useful information related to the current output in the input data, thereby improving the quality of the output, and the increasing attention module will bring continuous performance improvement.
  • the residual dual attention mechanism model uses a combination of spatial attention and channel attention in parallel, so that different types of effective information about water content in the gas chromatogram of the electronic grade hexafluorobutadiene to be measured are captured in large quantities, which can effectively enhance the feature recognition learning ability.
  • the enhanced gas chromatography characteristic graph is used as a decoding characteristic graph to perform decoding regression in a decoder to obtain a decoding value for representing the water content in the electronic-grade hexafluorobutadiene to be measured.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently detected, and then the quality inspection of the electronic-grade hexafluorobutadiene can be performed.
  • the residual dual attention mechanism model obtains an enhanced gas chromatography feature map as a decoding feature map by fusing deep and shallow features, namely, the weighted feature map F′ and the gas chromatography feature map F, and since the weighted feature map F′ as a deep feature is obtained by the channel attention mechanism and the spatial attention mechanism respectively, the feature distribution of the weighted feature map F′ conforms to the Gaussian distribution in the natural state, that is, the local feature distribution with average attention has the highest probability density, while the probability density of the local feature distribution with relatively high and relatively low attention is relatively low. In this way, the obtained weighted feature map F′ may have a poor clustering effect on the global correlation between the feature values, thereby weakening the key global feature distribution that affects the decoding result and affecting the decoding accuracy of the decoding feature map.
  • the weighted feature map F′ is subjected to feature clustering defocusing fuzzy optimization, which is expressed as:
  • ⁇ and ⁇ are the mean and standard deviation of the feature set fi,j, k ⁇ F′, respectively, and fi ,j,k is the feature value of the (i,j,k)th position of the weighted feature map F′.
  • the defocusing fuzzy optimization of the feature clustering compensates for the dependency similarity of the high-frequency distribution features following the Gaussian point distribution relative to the homogenized representation of the overall feature distribution by indexing the focus stack representation used to estimate the clustering metric value based on statistical information, thereby avoiding the focus fuzziness of the overall feature distribution caused by the low dependency similarity.
  • the feature clustering effect of the weighted feature graph F′ is improved, and the decoding accuracy of the decoding feature graph obtained by further fusing the weighted feature graph F′ and the gas chromatography feature graph F is optimized.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of the electronic-grade hexafluorobutadiene.
  • the present application provides a quality inspection system for electronic-grade hexafluorobutadiene, which includes: a gas chromatogram acquisition module, used to obtain a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured; a feature detection module, used to pass the gas chromatogram through a first convolutional neural network model including a significant detector to obtain a gas chromatogram feature graph; a feature enhancement module, used to pass the gas chromatogram feature graph through a residual double attention mechanism model to obtain an enhanced gas chromatogram feature graph as a decoding feature graph; and a decoding module, used to pass the decoded feature graph through a decoder to obtain a decoding value, wherein the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured.
  • a gas chromatogram acquisition module used to obtain a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured
  • a feature detection module used
  • FIG1 illustrates an application scenario diagram of a quality inspection system and method for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • electronic-grade hexafluorobutadiene to be measured e.g., H as shown in FIG1
  • a gas sampling pipe e.g., G as shown in FIG1
  • a gas chromatograph of the electronic-grade hexafluorobutadiene to be measured is collected by a gas chromatograph deployed next to the gas sampling pipe (e.g., T as shown in FIG1 ).
  • the collected gas chromatogram of the electronic-grade hexafluorobutadiene to be measured is input into a server (e.g., S as shown in FIG1 ) in which a quality inspection algorithm for electronic-grade hexafluorobutadiene is deployed, wherein the server can process the gas chromatogram of the electronic-grade hexafluorobutadiene to be measured using the quality inspection algorithm for the electronic-grade hexafluorobutadiene to generate a detection result for representing the water content in the electronic-grade hexafluorobutadiene to be measured.
  • a server e.g., S as shown in FIG1
  • the server can process the gas chromatogram of the electronic-grade hexafluorobutadiene to be measured using the quality inspection algorithm for the electronic-grade hexafluorobutadiene to generate a detection result for representing the water content in the electronic-grade hexafluorobutadiene to be measured.
  • Figure 2 illustrates a schematic block diagram of a quality inspection system for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • the quality inspection system 100 for electronic-grade hexafluorobutadiene includes: a gas chromatogram acquisition module 110 for acquiring a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured; a feature detection module 120 for passing the gas chromatogram through a first convolutional neural network model including a significant detector to obtain a gas chromatogram feature graph; a feature enhancement module 130 for passing the gas chromatogram feature graph through a residual dual attention mechanism model to obtain an enhanced gas chromatogram feature graph as a decoding feature graph; and a decoding module 140 for passing the decoded feature graph through a decoder to obtain a decoding value, wherein the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured.
  • the three methods for determining trace moisture in gas in the current national standard are not suitable for the electronic-grade hexafluorobutadiene quality inspection of the present application.
  • the chromatogram refers to the image of the detection signal of the separated component distributed over time, therefore, the water content of the electronic-grade hexafluorobutadiene can be obtained from the gas chromatogram, but because water only accounts for a small part in the electronic-grade hexafluorobutadiene, it is difficult to obtain an accurate water content measurement value through the gas chromatogram.
  • an artificial intelligence algorithm based on deep learning is used to extract multi-scale implicit feature distribution information of different sizes in the gas chromatogram of electronic grade hexafluorobutadiene, and an attention mechanism is used to further focus on the small-size water content detection in the electronic grade hexafluorobutadiene and filter out useless interfering feature information.
  • the water content in the electronic grade hexafluorobutadiene can be detected intelligently and accurately to ensure the product quality of the electronic grade hexafluorobutadiene.
  • the electronic grade hexafluorobutadiene to be measured is collected through a gas sampling pipe, and a gas chromatogram of the electronic grade hexafluorobutadiene to be measured is collected through a gas chromatograph deployed next to the gas sampling pipe.
  • the network structure of the standard convolutional neural network is adjusted, specifically, the first convolutional neural network model including a significant detector is used to encode the gas chromatogram to obtain a gas chromatogram feature map.
  • each layer of the significant detector uses a combination of large and small convolutional kernels to enable each layer of the convolutional neural network model to pay more attention to small-sized object features.
  • each layer of the convolutional neural network uses a two-dimensional convolution kernel to perform a convolution encoding on the input data. Accordingly, in the technical solution of the present application, in order to improve the ability of each layer of the convolutional neural network to extract local significant features, each layer of the convolutional neural network is transformed to perform secondary convolution, and the size of the first convolution kernel is larger than the size of the second convolution kernel.
  • convolution through a larger convolution kernel has a larger receptive field, but the extracted image feature pattern is rough, and it is easy to ignore the details with resolution in the gas chromatogram, such as the implicit features of moisture; accordingly, using a smaller convolution kernel for secondary convolution can better model local information, and thus can focus on more detailed feature information.
  • the feature detection module is further used to: use each layer of the first convolutional neural network model to perform the following on the input data in the forward pass of the layer: use a first convolution kernel to convolve the input data to obtain a first convolutional feature map; use a second convolution kernel to convolve the first convolutional feature map to obtain a second convolutional feature map, wherein the size of the first convolution kernel is larger than the size of the second convolution kernel; perform pooling on the second convolutional feature map to obtain a pooled feature map; and, perform activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolutional neural network model is the gas chromatogram feature map, and the input of the first layer of the first convolutional neural network model is the gas chromatogram.
  • the size of the first convolution kernel is 5 ⁇ 5
  • the size of the second convolution kernel is 3 ⁇ 3.
  • the feature enhancement module 130 is used to obtain an enhanced gas chromatogram feature graph as a decoding feature graph by passing the gas chromatogram feature graph through a residual dual attention mechanism model. It should be understood that, considering that the moisture content in the gas chromatogram belongs to a small-sized object (i.e., it accounts for a small proportion of the entire image), and secondly, there are other useless features and environmental factors in the gas chromatogram, which will interfere with the target detection of moisture content. Therefore, in the technical solution of the present application, the gas chromatogram feature graph is further enhanced in feature data to obtain an enhanced gas chromatogram feature graph.
  • the network will obtain partial feature information, but will not automatically distinguish the detailed information between high and low frequencies and the differences between features of each category.
  • the network's ability to selectively use features is limited.
  • the attention mechanism can select the focus position and produce a more discriminative feature representation, and the features after the attention module is added will undergo adaptive changes as the network deepens. Therefore, in the technical solution of the present application, a channel attention mechanism and a spatial attention mechanism are introduced, and by introducing a residual structure, they are combined with the proposed dual attention network to construct a residual dual attention model.
  • This model combines spatial attention and channel attention in parallel, so that different types of effective information are captured in large quantities, which can effectively enhance the feature recognition learning ability.
  • the task processing system is more focused on finding useful information in the input data that is significantly related to the current output, thereby improving the quality of the output, and the increasing attention modules will bring continuous performance improvements.
  • the residual dual attention mechanism model By using a combination of spatial attention and channel attention in parallel, a large amount of effective information about the water content in the gas chromatograms of different types of electronic-grade hexafluorobutadiene to be measured is captured, which can effectively enhance the feature recognition learning ability.
  • Figure 3 illustrates a block diagram of a feature enhancement module in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • the feature enhancement module 130 includes: a spatial attention unit 131, which is used to input the gas chromatographic feature map into the spatial attention module of the residual dual attention mechanism model to obtain a spatial attention map; a channel attention unit 132, which is used to input the gas chromatographic feature map into the channel attention module of the residual dual attention mechanism model to obtain a channel attention map; an attention fusion unit 133, which is used to fuse the spatial attention map and the channel attention map to obtain a fused attention map; an activation unit 134, which is used to input the fused attention map into a Sigmoid activation function for activation to obtain a fused attention feature map; an attention application unit 135, which is used to calculate the multiplication of the fused attention feature map and the gas chromatographic feature map by position point to obtain a weighted
  • FIG4 illustrates a block diagram of a spatial attention unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • the spatial attention unit 131 includes: a spatial perception subunit 1311, which is used to convolutionally encode the gas chromatographic characteristic graph using the convolution layer of the spatial attention module of the residual dual attention mechanism model to obtain a convolutional characteristic graph; a probabilistic subunit 1312, which is used to pass the spatial attention graph through a Softmax function to obtain a spatial attention score graph; and a spatial attention application subunit 1313, which is used to multiply the spatial attention score graph by the gas chromatographic characteristic graph by position point to obtain a spatial attention graph.
  • Figure 5 illustrates a block diagram of a channel attention unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • the channel attention unit 132 includes: a pooling subunit 1321 along the channel dimension, which is used to perform global mean pooling along the channel dimension on the gas chromatographic characteristic graph to obtain a channel feature vector; a nonlinear activation subunit 1322, which is used to pass the channel feature vector through a Softmax activation function to obtain an activated channel feature vector; and a channel attention application subunit 1323, which is used to weight the feature matrix along the channel dimension of the gas chromatographic characteristic graph using the eigenvalues of each position in the normalized channel feature vector as weights to obtain a channel attention graph.
  • the residual dual attention mechanism model obtains an enhanced gas chromatography feature map as a decoding feature map by fusing deep and shallow features, namely, the weighted feature map F′ and the gas chromatography feature map F, and since the weighted feature map F′ as a deep feature is obtained by the channel attention mechanism and the spatial attention mechanism respectively, the feature distribution of the weighted feature map F′ conforms to the Gaussian distribution in the natural state, that is, the local feature distribution with average attention has the highest probability density, while the probability density of the local feature distribution with relatively high and relatively low attention is relatively low.
  • the obtained weighted feature map F′ may have a poor clustering effect on the global correlation between the feature values, thereby weakening the key global feature distribution that affects the decoding result and affecting the decoding accuracy of the decoding feature map.
  • the weighted feature map F′ is subjected to defocusing fuzzy optimization of feature clustering.
  • FIG6 illustrates a block diagram of a residual fusion unit in a quality inspection system for electronic grade hexafluorobutadiene according to an embodiment of the present application.
  • the residual fusion unit 136 includes: a feature optimization subunit 1361, which is used to perform defocusing fuzzy optimization of feature clustering on the weighted feature map to obtain an optimized weighted feature map according to the following formula, wherein the formula is:
  • fi,j,k represents the eigenvalue of the (i,j,k)th position of the weighted feature graph
  • ⁇ and ⁇ respectively represent the mean and standard deviation of the set of eigenvalues at each position of the weighted feature graph
  • a fusion subunit 1362 is used to calculate the positionally weighted sum of the optimized weighted feature graph and the gas chromatography feature graph to obtain the enhanced gas chromatography feature graph.
  • the defocused blur optimization of feature clustering is performed by representing the focused stack used to estimate the clustering metric
  • the feature clustering index based on statistical information is performed to compensate for the dependency similarity of the high-frequency distribution features following the Gaussian point distribution relative to the homogenized representation of the overall feature distribution, thereby avoiding the focus blur of the overall feature distribution caused by the low dependency similarity.
  • the feature clustering effect of the weighted feature graph F' is improved, and the decoding accuracy of the decoding feature graph obtained by further fusing the weighted feature graph F' and the gas chromatography feature graph F is optimized.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of the electronic-grade hexafluorobutadiene.
  • the decoding module 140 is used to pass the decoding characteristic graph through a decoder to obtain a decoding value, and the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured. That is, after further feature data enhancement of the gas chromatographic characteristic graph, it is imported as the decoding characteristic graph for decoding regression to obtain a decoding value for representing the water content in the electronic-grade hexafluorobutadiene to be measured, so that the water content in the electronic-grade hexafluorobutadiene can be intelligently detected, and then the quality inspection of the electronic-grade hexafluorobutadiene can be performed.
  • the water content of the electronic-grade hexafluorobutadiene to be measured is determined, it is compared with the water content of the standard electronic-grade hexafluorobutadiene to determine whether the quality of the electronic-grade hexafluorobutadiene to be measured is qualified.
  • the decoding module is further used to: use the decoder to perform decoding regression on the decoding feature map to obtain the decoding value using the following formula, wherein the formula is Among them, X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.
  • the quality inspection system of the electronic-grade hexafluorobutadiene uses an artificial intelligence algorithm based on deep learning to extract multi-scale implicit feature distribution information of different sizes in the gas chromatogram of the electronic-grade hexafluorobutadiene, and uses an attention mechanism to further focus on the small-size water content detection in the electronic-grade hexafluorobutadiene and filter out useless interference feature information.
  • the water content in the electronic-grade hexafluorobutadiene can be intelligently and accurately detected to ensure the product quality of the electronic-grade hexafluorobutadiene.
  • the quality inspection system 100 for electronic-grade hexafluorobutadiene can be implemented in various terminal devices, such as a server with a quality inspection algorithm for electronic-grade hexafluorobutadiene deployed.
  • the quality inspection system 100 for electronic-grade hexafluorobutadiene can be integrated into the terminal device as a software module and/or a hardware module.
  • the quality inspection system 100 for electronic-grade hexafluorobutadiene can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the quality inspection system 100 for electronic-grade hexafluorobutadiene can also be one of the many hardware modules of the terminal device.
  • the electronic grade hexafluorobutadiene quality inspection system 100 and the terminal device may also be separate devices, and the electronic grade hexafluorobutadiene quality inspection system 100 may be connected to the terminal device via a wired and/or wireless network and transmit interactive information in accordance with an agreed data format.
  • FIG7 illustrates a flow chart of a quality inspection method for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • the quality inspection method for electronic-grade hexafluorobutadiene according to an embodiment of the present application includes: S110, obtaining a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured; S120, passing the gas chromatogram through a first convolutional neural network model including a significant detector to obtain a gas chromatogram characteristic graph; S130, passing the gas chromatogram characteristic graph through a residual dual attention mechanism model to obtain an enhanced gas chromatogram characteristic graph as a decoding characteristic graph; and, S140, passing the decoded characteristic graph through a decoder to obtain a decoding value, the decoding value being the water content in the electronic-grade hexafluorobutadiene to be measured.
  • Figure 8 illustrates a schematic diagram of the system architecture of the quality inspection system for electronic-grade hexafluorobutadiene according to an embodiment of the present application.
  • a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured is obtained.
  • the gas chromatogram is passed through a first convolutional neural network model including a significant detector to obtain a gas chromatogram characteristic graph.
  • the gas chromatogram characteristic graph is passed through a residual dual attention mechanism model to obtain an enhanced gas chromatogram characteristic graph as a decoding characteristic graph.
  • the decoded characteristic graph is passed through a decoder to obtain a decoding value, and the decoding value is the water content in the electronic-grade hexafluorobutadiene to be measured.
  • the gas chromatogram is passed through a first convolutional neural network model including a significant detector to obtain a gas chromatogram feature block, comprising: using the first convolutional neural network model In the forward pass of the layer, each layer respectively performs the following operations on the input data: using a first convolution kernel to perform convolution processing on the input data to obtain a first convolution feature map; using a second convolution kernel to perform convolution processing on the first convolution feature map to obtain a second convolution feature map, wherein the size of the first convolution kernel is larger than the size of the second convolution kernel; performing pooling processing on the second convolution feature map to obtain a pooled feature map; and, performing activation processing on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolution neural network model is the gas chromatogram feature map, and the input of the first layer of the first convolution neural network model is the gas chromatogram.
  • the size of the first convolution kernel is 5 ⁇ 5
  • the size of the second convolution kernel is 3 ⁇ 3.
  • FIG. 9 illustrates a flow chart of a method for quality inspection of electronic grade hexafluorobutadiene according to an embodiment of the present application, in which the gas chromatographic characteristic graph is subjected to a residual dual attention mechanism model to obtain an enhanced gas chromatographic characteristic graph as a decoding characteristic graph.
  • the gas chromatography characteristic graph is passed through the residual dual attention mechanism model to obtain an enhanced gas chromatography characteristic graph as a decoding characteristic graph, including: S210, inputting the gas chromatography characteristic graph into the spatial attention module of the residual dual attention mechanism model to obtain a spatial attention graph; S220, inputting the gas chromatography characteristic graph into the channel attention module of the residual dual attention mechanism model to obtain a channel attention graph; S230, fusing the spatial attention graph and the channel attention graph to obtain a fused attention graph; S240, inputting the fused attention graph into a Sigmoid activation function for activation to obtain a fused attention feature graph; S250, calculating the point-by-point multiplication of the fused attention feature graph and the gas chromatography characteristic graph to obtain a weighted feature graph; and, S260, fusing the weighted feature graph and the gas chromatography characteristic graph to obtain the enhanced gas chromatography feature graph.
  • the gas chromatogram characteristic graph is input into the spatial attention module of the residual dual attention mechanism model to obtain a spatial attention graph, comprising: using the convolution layer of the spatial attention module of the residual dual attention mechanism model to perform convolution encoding on the gas chromatogram characteristic graph to obtain a convolution feature graph; passing the spatial attention graph through a Softmax function to obtain a spatial attention score graph; and performing positional point multiplication of the spatial attention score graph and the gas chromatogram characteristic graph to obtain a spatial attention graph.
  • the gas chromatography feature map is input into the channel attention module of the residual dual attention mechanism model to obtain a channel attention map, including: performing global mean pooling on the gas chromatography feature map along the channel dimension to obtain a channel feature vector; passing the channel feature vector through a Softmax activation function to obtain an activated channel feature vector; and weighting the feature matrix of the gas chromatography feature map along the channel dimension using the eigenvalues of each position in the normalized channel feature vector as weights to obtain a channel attention map.
  • the fusion of the weighted characteristic graph and the gas chromatography characteristic graph to obtain the enhanced gas chromatography characteristic graph includes: performing defocusing fuzzy optimization of feature clustering on the weighted characteristic graph according to the following formula to obtain an optimized weighted characteristic graph, wherein the formula is:
  • fi,j,k represents the eigenvalue of the (i,j,k)th position of the weighted characteristic graph
  • ⁇ and ⁇ respectively represent the mean and standard deviation of the set of eigenvalues at each position of the weighted characteristic graph
  • the positionally weighted sum of the optimized weighted characteristic graph and the gas chromatography characteristic graph is calculated to obtain the enhanced gas chromatography characteristic graph.
  • the decoding feature map is passed through a decoder to obtain decoding, including: using the decoder to decode and regress the decoding feature map using the following formula to obtain the decoding value, using the following formula, wherein the formula is Where X is the decoded feature map, Y is the decoded value, and W is the weight matrix. Represents matrix multiplication.
  • FIG. 10 illustrates a block diagram of an electronic device according to an embodiment of the present application.
  • the electronic device 10 includes one or more processors 11 and a memory 12 .
  • the processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.
  • CPU central processing unit
  • the memory 12 may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory.
  • the volatile memory may include, for example, random access memory (RAM) and/or cache memory (cache), etc.
  • the non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc.
  • One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the quality inspection of the electronic-grade hexafluorobutadiene of each embodiment of the present application described above and/or other desired functions.
  • Various contents such as a gas chromatogram of the electronic-grade hexafluorobutadiene to be measured may also be stored in the computer-readable storage medium.
  • the electronic device 10 may further include: an input device 13 and an output device 14, and these components are interconnected via a bus system and/or other forms of connection mechanisms (not shown).
  • the input device 13 may include, for example, a keyboard, a mouse, etc.
  • the output device 14 can output various information to the outside, including the water content in the electronic grade hexafluorobutadiene to be measured, etc.
  • the output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.
  • an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the quality inspection method for electronic grade hexafluorobutadiene according to various embodiments of the present application described in the above-mentioned "Exemplary Methods" section of this specification.
  • the computer program product may be written in any combination of one or more programming languages to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages.
  • the program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.
  • an embodiment of the present application may also be a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, enables the processor to execute the steps of the quality inspection method for electronic grade hexafluorobutadiene according to various embodiments of the present application described in the above “Exemplary Method” section of this specification.
  • the computer readable storage medium can adopt any combination of one or more readable media.
  • the readable medium can be a readable signal medium or a readable storage medium.
  • the readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above.
  • readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
  • RAM random access memory
  • ROM read-only memory
  • EPROM or flash memory erasable programmable read-only memory
  • CD-ROM compact disk read-only memory
  • magnetic storage device or any suitable combination of the above.
  • each component or each step can be decomposed and/or recombined.
  • Such decomposition and/or recombination should be regarded as equivalent solutions of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Biochemistry (AREA)
  • Pathology (AREA)
  • Immunology (AREA)
  • Analytical Chemistry (AREA)
  • Artificial Intelligence (AREA)
  • Chemical & Material Sciences (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Other Investigation Or Analysis Of Materials By Electrical Means (AREA)

Abstract

一种电子级六氟丁二烯的质检系统（100）及其方法，具体的，首先，通过气相色谱仪（T）获取待测量电子级六氟丁二烯（H）的气相色谱图（S110）；然后，将气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图（S120）；接着，将气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图（S130）；最后，将解码特征图通过解码器以得到解码值，解码值为待测量电子级六氟丁二烯中含水量（S140），通过这样的方式，能够对于电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。

Description

一种电子级六氟丁二烯的质检系统及其方法 技术领域
本申请涉及气体质检技术领域，且更为具体地，涉及一种电子级六氟丁二烯的质检系统及其方法。
背景技术
电子级六氟丁二烯是一种刻蚀性能优良，绿色环保环境友好型干刻蚀气体，广泛应用于大规模集成电路和高速高容量存储芯片的生产过程中。其中，电子级六氟丁二烯中水分指标对于集成电路和芯片工艺的优化与提升带来了严重影响，因此，建立可靠的电子级六氟丁二烯中水分的分析技术是保证电子级六氟丁二烯产品质量的关键。
目前，现行国标中测定气体中微量水分的方法有三种：露点法、电解法和光腔衰荡法。其中，露点法操作简单，但是电子级六氟丁二烯会与氧化铝发生反应，损坏仪器，因此露点法不能满足要求。对于电解法和光腔衰荡法，由于电子级六氟丁二烯的沸点为6℃，很容易液化从而污染氢火焰离子化检测器，因此电解法和光腔衰荡法不适用。
因此，期待一种优化的用于电子级六氟丁二烯的质检方案。
发明内容
为了解决上述技术问题，提出了本申请。本申请的实施例提供了一种电子级六氟丁二烯的质检系统及其方法，其采用基于深度学习的人工智能算法来提取出电子级六氟丁二烯的气相色谱图中的不同尺寸大小的多尺度隐含特征分布信息，并且利用了注意力机制来进一步聚焦于所述电子级六氟丁二烯中的小尺寸含水量检测而滤除无用的干扰特征信息。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
根据本申请的一个方面，提供了一种电子级六氟丁二烯的质检系统，其包括：气相色谱图采集模块，用于获取待测量电子级六氟丁二烯的气相色谱图；特征检测模块，用于将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；特征增强模块，用于将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及解码模块，用于将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
根据本申请的另一方面，提供了一种电子级六氟丁二烯的质检方法，其包括：获取待测量电子级六氟丁二烯的气相色谱图；将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
根据本申请的再一方面，提供了一种电子设备，包括：处理器；以及，存储器，在所述存储器中存储有计算机程序指令，所述计算机程序指令在被所述处理器运行时使得所述处理器执行如上所述的电子级六氟丁二烯的质检方法。
根据本申请的又一方面，提供了一种计算机可读介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行如上所述的电子级六氟丁二烯的质检方法。
与现有技术相比，本申请提供的一种电子级六氟丁二烯的质检系统及其方法，其采用基于深度学习的人工智能算法来提取出电子级六氟丁二烯的气相色谱图中的不同尺寸大小的多尺度隐含特征分布信息，并且利用了注意力机制来进一步聚焦于所述电子级六氟丁二烯中的小尺寸含水量检测而滤除无用的干扰特征信息。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
附图说明
通过结合附图对本申请实施例进行更详细的描述，本申请的上述以及其他目的、特 征和优势将变得更加明显。附图用来提供对本申请实施例的进一步理解，并且构成说明书的一部分，与本申请实施例一起用于解释本申请，并不构成对本申请的限制。在附图中，相同的参考标号通常代表相同部件或步骤。
图1图示了根据本申请实施例的电子级六氟丁二烯的质检系统及其方法的应用场景图。
图2图示了根据本申请实施例的电子级六氟丁二烯的质检系统的框图示意图。
图3图示了根据本申请实施例的电子级六氟丁二烯的质检系统中特征增强模块的框图。
图4图示了根据本申请实施例的电子级六氟丁二烯的质检系统中空间注意力单元的框图。
图5图示了根据本申请实施例的电子级六氟丁二烯的质检系统中通道注意力单元的框图。
图6图示了根据本申请实施例的电子级六氟丁二烯的质检系统中残差融合单元的框图。
图7图示了根据本申请实施例的电子级六氟丁二烯的质检方法的流程图。
图8图示了根据本申请实施例的电子级六氟丁二烯的质检方法的系统架构的示意图。
图9图示了根据本申请实施例的电子级六氟丁二烯的质检方法中，将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图的流程图。
图10图示了根据本申请实施例的电子设备的框图。
具体实施方式
下面，将参考附图详细地描述根据本申请的示例实施例。显然，所描述的实施例仅仅是本申请的一部分实施例，而不是本申请的全部实施例，应理解，本申请不受这里描述的示例实施例的限制。
场景概述：
如上所述，电子级六氟丁二烯是一种刻蚀性能优良，绿色环保环境友好型干刻蚀气体，广泛应用于大规模集成电路和高速高容量存储芯片的生产过程中。其中，电子级六氟丁二烯中水分指标对于集成电路和芯片工艺的优化与提升带来了严重影响，因此，建立可靠的电子级六氟丁二烯中水分的分析技术是保证电子级六氟丁二烯产品质量的关键。
目前，现行国标中测定气体中微量水分的方法有三种：露点法、电解法和光腔衰荡法。其中，露点法操作简单，但是电子级六氟丁二烯会与氧化铝发生反应，损坏仪器，因此露点法不能满足要求。对于电解法和光腔衰荡法，由于电子级六氟丁二烯的沸点为6℃，很容易液化从而污染氢火焰离子化检测器，因此电解法和光腔衰荡法不适用。因此，期待一种优化的用于电子级六氟丁二烯的质检方案。
目前，深度学习以及神经网络已经广泛应用于计算机视觉、自然语言处理、语音信号处理等领域。此外，深度学习以及神经网络在图像分类、物体检测、语义分割、文本翻译等领域，也展现出了接近甚至超越人类的水平。
近年来，深度学习以及神经网络的发展为用于电子级六氟丁二烯质检的含水量检测提供了新的解决思路和方案。
相应地，考虑到由于色谱图是指被分离组分的检测信号随时间分布的图象，因此，从气相色谱图可以得出电子级六氟丁二烯的含水量，但是因水在电子级六氟丁二烯仅占微小部分，很难通过气相色谱图得到精准的含水量测量值。基于此，在本申请的技术方案中，采用基于深度学习的人工智能算法来提取出电子级六氟丁二烯的气相色谱图中的不同尺寸大小的多尺度隐含特征分布信息，并且利用了注意力机制来进一步聚焦于所述电子级六氟丁二烯中的小尺寸含水量检测而滤除无用的干扰特征信息。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
具体地，在本申请的技术方案中，首先，获取待测量电子级六氟丁二烯的气相色谱图。接着，使用在图像的隐含特征提取方面具有优异表现的卷积神经网络模型来进行所述待测量电子级六氟丁二烯的气相色谱图的特征挖掘，但是，考虑到卷积神经网络对图像中的小尺度特征(例如，本申请实施例中的待测量电子级六氟丁二烯的水量)的分辨能力不强。因此，在本申请的技术方案中，对标准卷积神经网络的网络结构进行调整，具体地，使用包含显著检测器的第一卷积神经网络模型来对于所述气相色谱图进行编码，以得到气相色谱特征图。这里，所述显著检测器的每一层使用大小卷积核的组合卷积核使得卷积神经网络模型的各层能够更加关注小尺寸的对象特征。
也就是说，在标准的卷积神经网络中，卷积神经网络的各层使用二维卷积核对输入数据进行一次卷积编码。相应地，在本申请的技术方案中，为了提高卷积神经网络的各层对局部显著特征的提取能力，将所述卷积神经网络的各层改造为进行二次卷积，并且，第一卷积核的尺寸大于第二卷积核的尺寸。应可以理解，通过尺寸较大的卷积核进行卷积具有较大的感受野，但其所提取的图像特征模式粗糙，容易忽略掉所述气相色谱图中具有分辨力的细节，例如水分的隐含特征；相应地，使用尺寸较小的卷积核进行二次卷积，可对局部信息进行更好地建模，进而能够关注到更加细节的特征信息。在本申请一个具体的示例中，所述第一卷积核的尺寸为5×5，所述第二卷积核的尺寸为3×3。
进一步地，在本申请的技术方案中，考虑到所述气相色谱图中的水分含量属于小尺寸对象(即，其在整个图像中所占比例较小)，其次，所述气相色谱图中还存在有其他无用特征以及环境因素的影响，这会对含水量的目标检测造成干扰。因此，在本申请的技术方案中，进一步对所述气相色谱特征图进行特征数据增强以得到增强气相色谱特征图。
具体地，网络经过一系列卷积之后，会得到部分特征信息，但不会自动区分高低频间的详细信息与各个类别特征间的差异性，网络选择性地使用特征的能力有限，鉴于注意力机制能够选择聚焦位置，产生更具分辨性的特征表示，且加入注意力模块后的特征会随着网络的加深产生适应性的改变。因此，在本申请的技术方案中，引入通道注意力机制和空间注意力机制，并通过引入残差结构，将其与提出的双注意力网络相结合来构造残差双注意力模型，此模型将空间注意力和通道注意力并行组合，使得不同类型的有效信息被大量捕捉到，可有效增强特征辨别学习能力，在网络训练过程中，任务处理系统更专注于找到输入数据中显著的与当前输出相关的有用信息，从而提高输出的质量，且渐增的注意力模块将带来持续的性能提升。应可以理解，所述残差双注意力机制模型通过使用空间注意力和通道注意力并行的组合，使得不同类型的所述待测量电子级六氟丁二烯的气相色谱图中的关于含水量的有效信息被大量捕捉到，可有效增强特征辨别学习能力。
然后，将所述增强气相色谱特征图作为解码特征图来通过解码器中进行解码回归，以得到用于表示所述待测量电子级六氟丁二烯中含水量的解码值。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能检测，进而进行电子级六氟丁二烯的质检。
特别地，在本申请的技术方案中，所述残差双注意力机制模型通过融合深层浅层特征，即加权特征图F′和所述气相色谱特征图F来得到增强气相色谱特征图作为解码特征图，并且，由于作为深层特征的加权特征图F′是分别通过通道注意力机制和空间注意力机制得到的，因此所述加权特征图F′的特征分布符合自然状态下的高斯分布，即获得平均注意力关注的局部特征分布具有最高的概率密度，而具有相对高和相对低的注意力关注的局部特征分布的概率密度都较低。这样，所获得的加权特征图F′可能存在各个特征值之间在全局关联关系上的聚类效果不良的情况，从而弱化了影响解码结果的关键全局特征分布，影响所述解码特征图的解码准确性。
基于此，对所述加权特征图F′进行特征聚类的去聚焦模糊优化，表示为：
μ和δ分别是特征集合fi，j，k∈F′的均值和标准差，且fi，j，k是所述加权特征图F′的第(i，j，k)位置的特征值。
这里，所述特征聚类的去聚焦模糊优化通过将用于估计聚类度量值的聚焦堆栈表示进行基于统计信息的特征聚类索引，来补偿遵循高斯点分布的高频分布特征相对于整体特征分布的均一化表示的依赖相似度，从而避免由于该依赖相似度低而引起整体特征分布的聚焦模糊，这样，就提升了所述加权特征图F′的特征聚类效果，优化了进一步融合所述加权特征图F′和所述气相色谱特征图F得到的解码特征图的解码准确性。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
基于此，本申请提供了一种电子级六氟丁二烯的质检系统，其包括：气相色谱图采集模块，用于获取待测量电子级六氟丁二烯的气相色谱图；特征检测模块，用于将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；特征增强模块，用于将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及，解码模块，用于将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
图1图示了根据本申请实施例的电子级六氟丁二烯的质检系统及其方法的应用场景图。如图1所示，在该应用场景中，通过采气管(例如，如图1中所示意的G)采集待测量电子级六氟丁二烯(例如，如图1中所示意的H)，并通过部署于采气管旁的气相色谱仪(例如，如图1中所示意的T)采集待测量电子级六氟丁二烯的气相色谱图。然后，将采集的所述待测量电子级六氟丁二烯的气相色谱图输入至部署有电子级六氟丁二烯的质检算法的服务器中(例如，图1中所示意的S)，其中，所述服务器能够使用所述电子级六氟丁二烯的质检算法对所述待测量电子级六氟丁二烯的气相色谱图进行处理以生成用于表示所述待测量电子级六氟丁二烯中含水量的检测结果。
在介绍了本申请的基本原理之后，下面将参考附图来具体介绍本申请的各种非限制性实施例。
示例性系统：
图2图示了根据本申请实施例的电子级六氟丁二烯的质检系统的框图示意图。如图2所示，根据本申请实施例的所述电子级六氟丁二烯的质检系统100，包括：气相色谱图采集模块110，用于获取待测量电子级六氟丁二烯的气相色谱图；特征检测模块120，用于将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；特征增强模块130，用于将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及，解码模块140，用于将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
在本申请实施例中，所述气相色谱图采集模块110，用于获取待测量电子级六氟丁二烯的气相色谱图。如上所述，考虑到电子级六氟丁二烯中水分指标对于集成电路和芯片工艺的优化与提升带来了严重影响，因此，建立可靠的电子级六氟丁二烯中水分的分析技术是保证电子级六氟丁二烯产品质量的关键。具体的，考虑到现行国标中测定气体中微量水分的三种方法(露点法、电解法和光腔衰荡法)都不适用于本申请的电子级六氟丁二烯质检。同时，考虑到由于色谱图是指被分离组分的检测信号随时间分布的图象，因此，从气相色谱图可以得出电子级六氟丁二烯的含水量，但是因水在电子级六氟丁二烯仅占微小部分，很难通过气相色谱图得到精准的含水量测量值。基于此，在本申请的技术方案中，采用基于深度学习的人工智能算法来提取出电子级六氟丁二烯的气相色谱图中的不同尺寸大小的多尺度隐含特征分布信息，并且利用了注意力机制来进一步聚焦于所述电子级六氟丁二烯中的小尺寸含水量检测而滤除无用的干扰特征信息。这样，能 够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
在本申请一个具体的实施例中，通过采气管采集待测量电子级六氟丁二烯，并通过部署于采气管旁的气相色谱仪采集待测量电子级六氟丁二烯的气相色谱图。
在本申请实施例中，所述特征检测模块120，用于将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图。应可以理解，考虑到卷积神经网络模型在图像的隐含特征提取方面具有优异表现，因此，在本申请的技术方案中，通过卷积神经网络模型来进行所述待测量电子级六氟丁二烯的气相色谱图的特征挖掘。进一步的，考虑到卷积神经网络对图像中的小尺度特征(例如，本申请实施例中的待测量电子级六氟丁二烯的水量)的分辨能力不强。因此，在本申请的技术方案中，对标准卷积神经网络的网络结构进行调整，具体地，使用包含显著检测器的第一卷积神经网络模型来对于所述气相色谱图进行编码，以得到气相色谱特征图。这里，所述显著检测器的每一层使用大小卷积核的组合卷积核使得卷积神经网络模型的各层能够更加关注小尺寸的对象特征。
也就是说，在标准的卷积神经网络中，卷积神经网络的各层使用二维卷积核对输入数据进行一次卷积编码。相应地，在本申请的技术方案中，为了提高卷积神经网络的各层对局部显著特征的提取能力，将所述卷积神经网络的各层改造为进行二次卷积，并且，第一卷积核的尺寸大于第二卷积核的尺寸。应可以理解，通过尺寸较大的卷积核进行卷积具有较大的感受野，但其所提取的图像特征模式粗糙，容易忽略掉所述气相色谱图中具有分辨力的细节，例如水分的隐含特征；相应地，使用尺寸较小的卷积核进行二次卷积，可对局部信息进行更好地建模，进而能够关注到更加细节的特征信息。
在本申请一个具体的实施例中，所述特征检测模块，进一步用于：使用所述第一卷积神经网络模型的各层在层的正向传递中分别对输入数据进行：使用第一卷积核对所述输入数据进行卷积处理以得到第一卷积特征图；使用第二卷积核对所述第一卷积特征图进行卷积处理以得到第二卷积特征图，其中，所述第一卷积核的尺寸大于所述第二卷积核的尺寸；对所述第二卷积特征图进行池化处理以得到池化特征图；以及，对所述池化特征图进行激活处理以得到激活特征图；其中，所述第一卷积神经网络模型的最后一层的输出为所述气相色谱特征图，所述第一卷积神经网络模型的第一层的输入为所述气相色谱图。
在本申请一个具体的实施例中，所述第一卷积核的尺寸为5×5，所述第二卷积核的尺寸为3×3。
在本申请实施例中，所述特征增强模块130，用于将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图。应可以理解，考虑到所述气相色谱图中的水分含量属于小尺寸对象(即，其在整个图像中所占比例较小)，其次，所述气相色谱图中还存在有其他无用特征以及环境因素的影响，这会对含水量的目标检测造成干扰。因此，在本申请的技术方案中，进一步对所述气相色谱特征图进行特征数据增强以得到增强气相色谱特征图。
具体地，网络经过一系列卷积之后，会得到部分特征信息，但不会自动区分高低频间的详细信息与各个类别特征间的差异性，网络选择性地使用特征的能力有限，鉴于注意力机制能够选择聚焦位置，产生更具分辨性的特征表示，且加入注意力模块后的特征会随着网络的加深产生适应性的改变。因此，在本申请的技术方案中，引入通道注意力机制和空间注意力机制，并通过引入残差结构，将其与提出的双注意力网络相结合来构造残差双注意力模型，此模型将空间注意力和通道注意力并行组合，使得不同类型的有效信息被大量捕捉到，可有效增强特征辨别学习能力，在网络训练过程中，任务处理系统更专注于找到输入数据中显著的与当前输出相关的有用信息，从而提高输出的质量，且渐增的注意力模块将带来持续的性能提升。应可以理解，所述残差双注意力机制模型 通过使用空间注意力和通道注意力并行的组合，使得不同类型的所述待测量电子级六氟丁二烯的气相色谱图中的关于含水量的有效信息被大量捕捉到，可有效增强特征辨别学习能力。
图3图示了根据本申请实施例的电子级六氟丁二烯的质检系统中特征增强模块的框图。如图3所示，在本申请一个具体的实施例中，所述特征增强模块130，包括：空间注意力单元131，用于将所述气相色谱特征图输入所述残差双注意力机制模型的空间注意力模块以得到空间注意力图；通道注意力单元132，用于将所述气相色谱特征图输入所述残差双注意力机制模型的通道注意力模块以得到通道注意力图；注意力融合单元133，用于融合所述空间注意力图和所述通道注意力图以得到融合注意力图；激活单元134，用于将所述融合注意力图输入Sigmoid激活函数进行激活以得到融合注意力特征图；注意力施加单元135，用于计算所述融合注意力特征图和所述气相色谱特征图的按位置点乘以得到加权特征图；以及，残差融合单元136，用于融合所述加权特征图和所述气相色谱特征图以得到所述增强气相色谱特征图。
图4图示了根据本申请实施例的电子级六氟丁二烯的质检系统中空间注意力单元的框图。如图4所示，在本申请一个具体的实施例中，所述空间注意力单元131，包括：空间感知子单元1311，用于使用所述残差双注意力机制模型的空间注意力模块的卷积层对所述气相色谱特征图进行卷积编码以得到卷积特征图；概率化子单元1312，用于将所述空间注意力图通过Softmax函数以得到空间注意力得分图；以及，空间注意力施加子单元1313，用于将所述空间注意力得分图与所述气相色谱特征图进行按位置点乘以得到空间注意力图。
图5图示了根据本申请实施例的电子级六氟丁二烯的质检系统中通道注意力单元的框图。如图5所示，在本申请一个具体的实施例中，所述通道注意力单元132，包括：沿通道维度池化子单元1321，用于对所述气相色谱特征图进行沿通道维度的全局均值池化以得到通道特征向量；非线性激活子单元1322，用于将所述通道特征向量通过Softmax激活函数以得到激活通道特征向量；以及，通道注意力施加子单元1323，用于以所述归一化通道特征向量中各个位置的特征值作为权重对所述气相色谱特征图的沿通道维度的特征矩阵进行加权以得到通道注意力图。
特别地，在本申请的技术方案中，所述残差双注意力机制模型通过融合深层浅层特征，即加权特征图F′和所述气相色谱特征图F来得到增强气相色谱特征图作为解码特征图，并且，由于作为深层特征的加权特征图F′是分别通过通道注意力机制和空间注意力机制得到的，因此所述加权特征图F′的特征分布符合自然状态下的高斯分布，即获得平均注意力关注的局部特征分布具有最高的概率密度，而具有相对高和相对低的注意力关注的局部特征分布的概率密度都较低。这样，所获得的加权特征图F′可能存在各个特征值之间在全局关联关系上的聚类效果不良的情况，从而弱化了影响解码结果的关键全局特征分布，影响所述解码特征图的解码准确性。基于此，对所述加权特征图F′进行特征聚类的去聚焦模糊优化。
图6图示了根据本申请实施例的电子级六氟丁二烯的质检系统中残差融合单元的框图。如图6所示，在本申请一个具体的实施例中，所述残差融合单元136，包括：特征优化子单元1361，用于以如下公式对所述加权特征图进行特征聚类的去聚焦模糊优化以得到优化加权特征图，其中，所述公式为：
其中fi，j，k表示所述加权特征图的第(i，j，k)位置的特征值，μ和δ分别表示所述加权特征图的各个位置的特征值集合的均值和标准差；以及，融合子单元1362，用于计算所述优化加权特征图和所述气相色谱特征图的按位置加权和以得到所述增强气相色谱特征图。
这里，所述特征聚类的去聚焦模糊优化通过将用于估计聚类度量值的聚焦堆栈表示 进行基于统计信息的特征聚类索引，来补偿遵循高斯点分布的高频分布特征相对于整体特征分布的均一化表示的依赖相似度，从而避免由于该依赖相似度低而引起整体特征分布的聚焦模糊，这样，就提升了所述加权特征图F′的特征聚类效果，优化了进一步融合所述加权特征图F′和所述气相色谱特征图F得到的解码特征图的解码准确性。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
在本申请实施例中，所述解码模块140，用于将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。也就是，在对所述气相色谱特征图进行进一步的特征数据增强后，将其作为所述解码特征图导入进行解码回归，以获得用于表示所述待测量电子级六氟丁二烯中含水量的解码值，这样，能够对于所述电子级六氟丁二烯中的含水量进行智能检测，进而进行电子级六氟丁二烯的质检。
进一步地，在确定好所述待测量电子级六氟丁二烯中含水量后，将其与标准电子级六氟丁二烯中含水量进行比较，以确定所述待测量电子级六氟丁二烯质量是否合格。
在本申请一个具体的实施例中，所述解码模块，进一步用于：使用所述解码器以如下公式对所述解码特征图进行解码回归以得到所述解码值，其中，所述公式为 其中，X是解码特征图，Y是解码值，W是权重矩阵，表示矩阵乘法。
综上，基于本申请实施例的所述电子级六氟丁二烯的质检系统，其采用基于深度学习的人工智能算法来提取出电子级六氟丁二烯的气相色谱图中的不同尺寸大小的多尺度隐含特征分布信息，并且利用了注意力机制来进一步聚焦于所述电子级六氟丁二烯中的小尺寸含水量检测而滤除无用的干扰特征信息。这样，能够对于所述电子级六氟丁二烯中的含水量进行智能准确地检测，以保证电子级六氟丁二烯的产品质量。
如上所述，根据本申请实施例的所述电子级六氟丁二烯的质检系统100可以实现在各种终端设备中，例如部署有电子级六氟丁二烯的质检算法的服务器等。在一个示例中，电子级六氟丁二烯的质检系统100可以作为一个软件模块和/或硬件模块而集成到终端设备中。例如，该电子级六氟丁二烯的质检系统100可以是该终端设备的操作系统中的一个软件模块，或者可以是针对于该终端设备所开发的一个应用程序；当然，该电子级六氟丁二烯的质检系统100同样可以是该终端设备的众多硬件模块之一。
替换地，在另一示例中，该电子级六氟丁二烯的质检系统100与该终端设备也可以是分立的设备，并且电子级六氟丁二烯的质检系统100可以通过有线和/或无线网络连接到该终端设备，并且按照约定的数据格式来传输交互信息。
示例性方法：图7图示了根据本申请实施例的电子级六氟丁二烯的质检方法的流程图。如图7所示，根据本申请实施例的所述电子级六氟丁二烯的质检方法，包括：S110，获取待测量电子级六氟丁二烯的气相色谱图；S120，将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；S130，将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及，S140，将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
图8图示了根据本申请实施例的电子级六氟丁二烯的质检系统的系统架构的示意图。如图8所示，在本申请实施例所述电子级六氟丁二烯的质检系统的系统架构中，首先，获取待测量电子级六氟丁二烯的气相色谱图。然后，将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图。接着，将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图。最后，将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
在本申请一个具体的实施例中，所述将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图块，包括：使用所述第一卷积神经网络模型的 各层在层的正向传递中分别对输入数据进行：使用第一卷积核对所述输入数据进行卷积处理以得到第一卷积特征图；使用第二卷积核对所述第一卷积特征图进行卷积处理以得到第二卷积特征图，其中，所述第一卷积核的尺寸大于所述第二卷积核的尺寸；对所述第二卷积特征图进行池化处理以得到池化特征图；以及，对所述池化特征图进行激活处理以得到激活特征图；其中，所述第一卷积神经网络模型的最后一层的输出为所述气相色谱特征图，所述第一卷积神经网络模型的第一层的输入为所述气相色谱图。
在本申请一个具体的实施例中，所述第一卷积核的尺寸为5×5，所述第二卷积核的尺寸为3×3。
图9图示了根据本申请实施例的电子级六氟丁二烯的质检方法中，将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图的流程图。如图9所示，在本申请一个具体的实施例中，所述将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图，包括：S210，将所述气相色谱特征图输入所述残差双注意力机制模型的空间注意力模块以得到空间注意力图；S220，将所述气相色谱特征图输入所述残差双注意力机制模型的通道注意力模块以得到通道注意力图；S230，融合所述空间注意力图和所述通道注意力图以得到融合注意力图；S240，将所述融合注意力图输入Sigmoid激活函数进行激活以得到融合注意力特征图；S250，计算所述融合注意力特征图和所述气相色谱特征图的按位置点乘以得到加权特征图；以及，S260，融合所述加权特征图和所述气相色谱特征图以得到所述增强气相色谱特征图。
在本申请一个具体的实施例中，所述将所述气相色谱特征图输入所述残差双注意力机制模型的空间注意力模块以得到空间注意力图，包括：使用所述残差双注意力机制模型的空间注意力模块的卷积层对所述气相色谱特征图进行卷积编码以得到卷积特征图；将所述空间注意力图通过Softmax函数以得到空间注意力得分图；以及，将所述空间注意力得分图与所述气相色谱特征图进行按位置点乘以得到空间注意力图。
在本申请一个具体的实施例中，所述将所述气相色谱特征图输入所述残差双注意力机制模型的通道注意力模块以得到通道注意力图，包括：对所述气相色谱特征图进行沿通道维度的全局均值池化以得到通道特征向量；将所述通道特征向量通过Softmax激活函数以得到激活通道特征向量；以及，以所述归一化通道特征向量中各个位置的特征值作为权重对所述气相色谱特征图的沿通道维度的特征矩阵进行加权以得到通道注意力图。
在本申请一个具体的实施例中，所述融合所述加权特征图和所述气相色谱特征图以得到所述增强气相色谱特征图，包括：以如下公式对所述加权特征图进行特征聚类的去聚焦模糊优化以得到优化加权特征图，其中，所述公式为：
其中fi，j，k表示所述加权特征图的第(i，j，k)位置的特征值，μ和δ分别表示所述加权特征图的各个位置的特征值集合的均值和标准差；以及，计算所述优化加权特征图和所述气相色谱特征图的按位置加权和以得到所述增强气相色谱特征图。
在本申请一个具体的实施例中，所述将所述解码特征图通过解码器以得到解码，包括：使用所述解码器以如下公式对所述解码特征图进行解码回归以得到所述解码值，以如下公式，其中，所述公式为其中X是解码特征图，Y是解码值，W是权重矩阵，表示矩阵乘法。
这里，本领域技术人员可以理解，上述电子级六氟丁二烯的质检方法中的各个步骤的具体功能和操作已经在上面参考图1到图6的电子级六氟丁二烯的质检系统的描述中得到了详细介绍，并因此，将省略其重复描述。
示例性电子设备：
下面，参考图10来描述根据本申请实施例的电子设备。
图10图示了根据本申请实施例的电子设备的框图。
如图10所示，电子设备10包括一个或多个处理器11和存储器12。
处理器11可以是中央处理单元(CPU)或者具有数据处理能力和/或指令执行能力的其他形式的处理单元，并且可以控制电子设备10中的其他组件以执行期望的功能。
存储器12可以包括一个或多个计算机程序产品，所述计算机程序产品可以包括各种形式的计算机可读存储介质，例如易失性存储器和/或非易失性存储器。所述易失性存储器例如可以包括随机存取存储器(RAM)和/或高速缓冲存储器(cache)等。所述非易失性存储器例如可以包括只读存储器(ROM)、硬盘、闪存等。在所述计算机可读存储介质上可以存储一个或多个计算机程序指令，处理器11可以运行所述程序指令，以实现上文所述的本申请的各个实施例的电子级六氟丁二烯的质检以及/或者其他期望的功能。在所述计算机可读存储介质中还可以存储诸如待测量电子级六氟丁二烯的气相色谱图等各种内容。
在一个示例中，电子设备10还可以包括：输入装置13和输出装置14，这些组件通过总线系统和/或其他形式的连接机构(未示出)互连。
该输入装置13可以包括例如键盘、鼠标等等。
该输出装置14可以向外部输出各种信息，包括待测量电子级六氟丁二烯中含水量等。该输出装置14可以包括例如显示器、扬声器、打印机、以及通信网络及其所连接的远程输出设备等等。
示例性计算机程序产品和计算机可读存储介质：
除了上述方法和设备以外，本申请的实施例还可以是计算机程序产品，其包括计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本申请各种实施例的电子级六氟丁二烯的质检方法的步骤。
所述计算机程序产品可以以一种或多种程序设计语言的任意组合来编写用于执行本申请实施例操作的程序代码，所述程序设计语言包括面向对象的程序设计语言，诸如Java、C++等，还包括常规的过程式程序设计语言，诸如“C”语言或类似的程序设计语言。程序代码可以完全地在用户计算设备上执行、部分地在用户设备上执行、作为一个独立的软件包执行、部分在用户计算设备上部分在远程计算设备上执行、或者完全在远程计算设备或服务器上执行。
此外，本申请的实施例还可以是计算机可读存储介质，其上存储有计算机程序指令，所述计算机程序指令在被处理器运行时使得所述处理器执行本说明书上述“示例性方法”部分中描述的根据本申请各种实施例的电子级六氟丁二烯的质检方法的步骤。
所述计算机可读存储介质可以采用一个或多个可读介质的任意组合。可读介质可以是可读信号介质或者可读存储介质。可读存储介质例如可以包括但不限于电、磁、光、电磁、红外线、或半导体的系统、装置或器件，或者任意以上的组合。可读存储介质的更具体的例子(非穷举的列表)包括：具有一个或多个导线的电连接、便携式盘、硬盘、随机存取存储器(RAM)、只读存储器(ROM)、可擦式可编程只读存储器(EPROM或闪存)、光纤、便携式紧凑盘只读存储器(CD-ROM)、光存储器件、磁存储器件、或者上述的任意合适的组合。
以上结合具体实施例描述了本申请的基本原理，但是，需要指出的是，在本申请中提及的优点、优势、效果等仅是示例而非限制，不能认为这些优点、优势、效果等是本申请的各个实施例必须具备的。另外，上述公开的具体细节仅是为了示例的作用和便于理解的作用，而非限制，上述细节并不限制本申请为必须采用上述具体的细节来实现。
本申请中涉及的器件、装置、设备、系统的方框图仅作为例示性的例子并且不意图要求或暗示必须按照方框图示出的方式进行连接、布置、配置。如本领域技术人员将认识到的，可以按任意方式连接、布置、配置这些器件、装置、设备、系统。诸如“包括”、“包含”、“具有”等等的词语是开放性词汇，指“包括但不限于”，且可与其互换使 用。这里所使用的词汇“或”和“和”指词汇“和/或”，且可与其互换使用，除非上下文明确指示不是如此。这里所使用的词汇“诸如”指词组“诸如但不限于”，且可与其互换使用。
还需要指出的是，在本申请的装置、设备和方法中，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本申请的等效方案。
提供所公开的方面的以上描述以使本领域的任何技术人员能够做出或者使用本申请。对这些方面的各种修改对于本领域技术人员而言是非常显而易见的，并且在此定义的一般原理可以应用于其他方面而不脱离本申请的范围。因此，本申请不意图被限制到在此示出的方面，而是按照与在此公开的原理和新颖的特征一致的最宽范围。
为了例示和描述的目的已经给出了以上描述。此外，此描述不意图将本申请的实施例限制到在此公开的形式。尽管以上已经讨论了多个示例方面和实施例，但是本领域技术人员将认识到其某些变型、修改、改变、添加和子组合。

Claims (10)

  1. 一种电子级六氟丁二烯的质检系统，其特征在于，包括：气相色谱图采集模块，用于获取待测量电子级六氟丁二烯的气相色谱图；特征检测模块，用于将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；特征增强模块，用于将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及解码模块，用于将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子级六氟丁二烯中含水量。
  2. 根据权利要求1所述的电子级六氟丁二烯的质检系统，其特征在于，所述特征检测模块，进一步用于：使用所述第一卷积神经网络模型的各层在层的正向传递中分别对输入数据进行：使用第一卷积核对所述输入数据进行卷积处理以得到第一卷积特征图；使用第二卷积核对所述第一卷积特征图进行卷积处理以得到第二卷积特征图，其中，所述第一卷积核的尺寸大于所述第二卷积核的尺寸；对所述第二卷积特征图进行池化处理以得到池化特征图；以及对所述池化特征图进行激活处理以得到激活特征图；其中，所述第一卷积神经网络模型的最后一层的输出为所述气相色谱特征图，所述第一卷积神经网络模型的第一层的输入为所述气相色谱图。
  3. 根据权利要求2所述的电子级六氟丁二烯的质检系统，其特征在于，所述第一卷积核的尺寸为5×5，所述第二卷积核的尺寸为3×3。
  4. 根据权利要求3所述的电子级六氟丁二烯的质检系统，其特征在于，所述特征增强模块，包括：空间注意力单元，用于将所述气相色谱特征图输入所述残差双注意力机制模型的空间注意力模块以得到空间注意力图；通道注意力单元，用于将所述气相色谱特征图输入所述残差双注意力机制模型的通道注意力模块以得到通道注意力图；注意力融合单元，用于融合所述空间注意力图和所述通道注意力图以得到融合注意力图；激活单元，用于将所述融合注意力图输入Sigmoid激活函数进行激活以得到融合注意力特征图；注意力施加单元，用于计算所述融合注意力特征图和所述气相色谱特征图的按位置点乘以得到加权特征图；以及残差融合单元，用于融合所述加权特征图和所述气相色谱特征图以得到所述增强气相色谱特征图。
  5. 根据权利要求4所述的电子级六氟丁二烯的质检系统，其特征在于，所述空间注意力单元，包括：空间感知子单元，用于使用所述残差双注意力机制模型的空间注意力模块的卷积层对所述气相色谱特征图进行卷积编码 以得到卷积特征图；概率化子单元，用于将所述空间注意力图通过Softmax函数以得到空间注意力得分图；以及空间注意力施加子单元，用于将所述空间注意力得分图与所述气相色谱特征图进行按位置点乘以得到所述空间注意力图。
  6. 根据权利要求5所述的电子级六氟丁二烯的质检系统，其特征在于，所述通道注意力单元，包括：沿通道维度池化子单元，用于对所述气相色谱特征图进行沿通道维度的全局均值池化以得到通道特征向量；非线性激活子单元，用于将所述通道特征向量通过Softmax激活函数以得到激活通道特征向量；以及通道注意力施加子单元，用于以所述归一化通道特征向量中各个位置的特征值作为权重对所述气相色谱特征图的沿通道维度的特征矩阵进行加权以得到所述通道注意力图。
  7. 根据权利要求6所述的电子级六氟丁二烯的质检系统，其特征在于，所述残差融合单元，包括：特征优化子单元，用于以如下公式对所述加权特征图进行特征聚类的去聚焦模糊优化以得到优化加权特征图，其中，所述公式为：
    其中fi，j，k表示所述加权特征图的第(i，j，k)位置的特征值，μ和δ分别表示所述加权特征图的各个位置的特征值集合的均值和标准差；以及融合子单元，用于计算所述优化加权特征图和所述气相色谱特征图的按位置加权和以得到所述增强气相色谱特征图。
  8. 根据权利要求7所述的电子级六氟丁二烯的质检系统，其特征在于，所述解码模块，进一步用于：使用所述解码器以如下公式对所述解码特征图进行解码回归以得到所述解码值，以如下公式，其中，所述公式为 其中X是解码特征图，Y是解码值，W是权重矩阵，表示矩阵乘法。
  9. 一种电子级六氟丁二烯的质检方法，其特征在于，包括：获取待测量电子级六氟丁二烯的气相色谱图；将所述气相色谱图通过包含显著检测器的第一卷积神经网络模型以得到气相色谱特征图；将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图；以及将所述解码特征图通过解码器以得到解码值，所述解码值为所述待测量电子 级六氟丁二烯中含水量。
  10. 根据权利要求9所述的电子级六氟丁二烯的质检方法，其特征在于，所述将所述气相色谱特征图通过残差双注意力机制模型以得到增强气相色谱特征图作为解码特征图，包括：将所述气相色谱特征图输入所述残差双注意力机制模型的空间注意力模块以得到空间注意力图；将所述气相色谱特征图输入所述残差双注意力机制模型的通道注意力模块以得到通道注意力图；融合所述空间注意力图和所述通道注意力图以得到融合注意力图；将所述融合注意力图输入Sigmoid激活函数进行激活以得到融合注意力特征图；计算所述融合注意力特征图和所述气相色谱特征图的按位置点乘以得到加权特征图；以及融合所述加权特征图和所述气相色谱特征图以得到所述增强气相色谱特征图。
PCT/CN2023/086973 2022-11-16 2023-04-07 一种电子级六氟丁二烯的质检系统及其方法 WO2024103613A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211437119.6 2022-11-16
CN202211437119.6A CN115932140A (zh) 2022-11-16 2022-11-16 一种电子级六氟丁二烯的质检系统及其方法

Publications (1)

Publication Number Publication Date
WO2024103613A1 true WO2024103613A1 (zh) 2024-05-23

Family

ID=86556704

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/086973 WO2024103613A1 (zh) 2022-11-16 2023-04-07 一种电子级六氟丁二烯的质检系统及其方法

Country Status (2)

Country Link
CN (1) CN115932140A (zh)
WO (1) WO2024103613A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116106457B (zh) * 2023-04-13 2023-06-27 天津海河标测技术检测有限公司 空气采样检测一体化装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113433250A (zh) * 2021-08-12 2021-09-24 中船重工（邯郸）派瑞特种气体有限公司 一种测定六氟丁二烯中水含量的装置及方法
CN114663821A (zh) * 2022-05-18 2022-06-24 武汉大学 基于视频高光谱成像技术的产品质量实时无损检测方法
US20220222914A1 (en) * 2021-01-14 2022-07-14 Tata Consultancy Services Limited System and method for attention-based surface crack segmentation
CN114815930A (zh) * 2022-06-30 2022-07-29 烟台黄金职业学院 煅烧器的温度控制系统及其温度控制方法
CN115346345A (zh) * 2022-07-28 2022-11-15 福建省杭氟电子材料有限公司 一种用于六氟丁二烯制备的智能化有毒有害气体报警系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220222914A1 (en) * 2021-01-14 2022-07-14 Tata Consultancy Services Limited System and method for attention-based surface crack segmentation
CN113433250A (zh) * 2021-08-12 2021-09-24 中船重工（邯郸）派瑞特种气体有限公司 一种测定六氟丁二烯中水含量的装置及方法
CN114663821A (zh) * 2022-05-18 2022-06-24 武汉大学 基于视频高光谱成像技术的产品质量实时无损检测方法
CN114815930A (zh) * 2022-06-30 2022-07-29 烟台黄金职业学院 煅烧器的温度控制系统及其温度控制方法
CN115346345A (zh) * 2022-07-28 2022-11-15 福建省杭氟电子材料有限公司 一种用于六氟丁二烯制备的智能化有毒有害气体报警系统

Also Published As

Publication number Publication date
CN115932140A (zh) 2023-04-07

Similar Documents

Publication Publication Date Title
CN115203380B (zh) 基于多模态数据融合的文本处理系统及其方法
Wang et al. Transfer learning with dynamic distribution adaptation
Chen et al. Good visual guidance makes a better extractor: Hierarchical visual prefix for multimodal entity and relation extraction
Li et al. CLIP-ReID: exploiting vision-language model for image re-identification without concrete text labels
WO2024113435A1 (zh) 基于大数据的电梯故障模式识别系统
Shen et al. DS-YOLOv8-Based object detection method for remote sensing images
Chen et al. Internvideo-ego4d: A pack of champion solutions to ego4d challenges
WO2024098653A1 (zh) 用于六氟磷酸锂制备的自动化采样分析系统及其方法
WO2024103613A1 (zh) 一种电子级六氟丁二烯的质检系统及其方法
CN115754108A (zh) 一种电子级六氟丁二烯的酸度测定系统及其方法
CN116912597A (zh) 知识产权智能管理系统及其方法
Sun et al. Part-based face recognition with vision transformers
Li et al. Sd4match: Learning to prompt stable diffusion model for semantic matching
Li et al. Adaptive uncertainty-based learning for text-based person retrieval
CN116467485B (zh) 一种视频图像检索构建系统及其方法
WO2024012289A1 (zh) 视频生成方法、装置、电子设备及介质
CN111460224A (zh) 评论数据的质量标注方法、装置、设备及存储介质
Dai et al. FMAP: Learning robust and accurate local feature matching with anchor points
Guo et al. One-stage low-resolution text recognition with high-resolution knowledge transfer
Sun et al. CLIP-driven Outliers Synthesis for few-shot OOD detection
Zhang et al. Psnd: A robust parking space number detector
CN117992835B (zh) 多策略标签消歧的偏多标签分类方法、设备及存储介质
Peng et al. Adaptive feature fusion for scene text script identification
Wu et al. Naster: non-local attentional scene text recognizer
CN118314161B (zh) 基于文本语义引导的小样本医学图像分割方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23890058

Country of ref document: EP

Kind code of ref document: A1