CN-122020258-A - VOCs leakage gas multi-mode data monitoring and data management method and system based on artificial intelligence

Abstract

The invention discloses a VOCs leakage gas multi-mode data monitoring and data management method and system based on artificial intelligence, wherein the method comprises the steps of preprocessing infrared monitoring images of a target scene and gas sensor data, and respectively inputting the preprocessed infrared monitoring images and gas sensor data into a first branch and a second branch for multi-scale feature extraction; the method comprises the steps of carrying out interactive enhancement on characteristic data from a first branch and a second branch through an interactive fusion Manba model to respectively obtain first data and second data, and carrying out linear projection on the first data and the second data after splicing to obtain fusion enhancement characteristics.

Inventors

• HE JINYONG
• LUO XIAOXIA
• CHEN YAN
• HUANG LUXIA
• LI RONGQING
• SHAO JINGXIAN
• GUO XIAOMA
• LIU QINGYUN
• LI SHUJIAN
• ZHONG JIANAN
• LI DONGYANG
• LIU LING
• LI JINHAO
• ZHANG YUEJUAN
• HU XIAORONG
• XU HUAIZHOU
• LI JIACONG
• SUN LIWANG
• LU YUHAN
• SHEN QIANHUI
• JIANG YU

Assignees

• 深圳深态环境科技有限公司
• 深圳市生态环境智能管控中心

Dates

Publication Date

20260512

Application Date

20260213

Claims (7)

1. 1. The utility model provides a VOCs leakage gas multimode data monitoring and data management method based on artificial intelligence, which is characterized by comprising the following steps: The method comprises the steps of acquiring infrared monitoring images and gas sensor data of a target scene, preprocessing the infrared monitoring images and the gas sensor data, and respectively inputting the preprocessed infrared monitoring images and the preprocessed gas sensor data into a first branch and a second branch for multi-scale feature extraction; The feature extraction of the first branch is based on a vision selective scanning state space block, and the feature extraction of the second branch is based on a selective scanning state space block; The feature data under each scale in the output of the first branch and the second branch are fused based on a first strategy to obtain a fusion enhancement feature set; The first strategy comprises the steps of carrying out interaction enhancement on characteristic data from a first branch and a second branch through an interaction fusion Manba model to respectively obtain first data and second data, and carrying out linear projection on the first data and the second data after splicing to obtain fusion enhancement characteristics.
2. 2. The method for monitoring and managing multi-modal data of VOCs leakage gas based on artificial intelligence according to claim 1, wherein said acquiring the infrared monitoring image of the target scene and the gas sensor data comprises: Acquiring infrared monitoring images ; Acquiring gas sensor data ; Wherein, the The height is H, the width is W, and the channel number is Is a matrix of (a); represents a matrix of height L and width d.
3. 3. The artificial intelligence based multi-modal data monitoring and data management method of VOCs leakage gas of claim 2, wherein the preprocessing comprises: the infrared monitoring image is standardized, which comprises the following steps: ; Wherein, the Representing the infrared monitoring image after the standardized treatment; representing an original infrared monitoring image; Representing the average value of the original infrared monitoring image; Representing standard deviation of the original infrared monitoring image; the standardized processing of the gas sensor data includes: ; Wherein, the Representing the normalized gas sensor data; representing the average value of each channel of the sensor data; representing standard deviation of each channel of sensor data; carrying out data enhancement on the standardized infrared monitoring image by adopting random rotation and color dithering; carrying out data enhancement on the standardized gas sensor data by adopting time sequence dithering; the infrared monitoring image after standardized processing and data enhancement is divided into local blocks by a Vision Transformer-style image blocking module and mapped to a high-dimensional feature space, and the method comprises the following steps: ; Wherein, the Representing the image data output by the image segmentation module, , Indicating the height is Wide as A matrix with the number of channels of C k , wherein k represents the scale level; Representing a clipping module; The enhanced gas sensor data is divided into local time sequence blocks, the gas sensor data is subjected to space long-range context feature extraction through a multi-scale selective scanning state space block and mapped to a high-dimensional feature space, and the method comprises the following steps: ; Wherein, the Representing the mapped high-dimensional features, , Indicating the height is Wide as K represents a scale level; Representing the time sequence partitioning and linear projection modules.
4. 4. The method for monitoring and managing multi-modal data of VOCs leakage gas based on artificial intelligence according to claim 3, wherein the interactively enhancing the feature data from the first branch and the second branch by interactively fusing the mannba model, respectively, comprises: Mapping the infrared monitoring image features from the first branch and the gas sensor data features from the second branch to uniform feature dimensions and eliminating scale differences through linear projection layer and layer normalization to obtain infrared intermediate quantity features after projection normalization And gas sensor intermediate quantity feature : ; ; Wherein, the A projection weight matrix for infrared monitoring image features, For the corresponding bias term(s), (-) Represents a layer normalization operation; a projected weight matrix that is characteristic of the gas sensor data, For the corresponding bias term(s), Representing the projection as a function identifier; Will be And The method comprises the steps of respectively inputting the first result and the second result into a selective scanning state space model block and a gating mechanism module to extract space context information and cross-modal enhancement information, wherein the step of respectively obtaining the first result and the second result comprises the following steps: ; ; ; Wherein, the , Representing the discretized system projection matrix; Representing an exponential function; an adjustment parameter representing a time step; representing a system state transition matrix; representing a system projection matrix; representing the discretized system projection matrix; Representing the hidden state of the time step t; Representing the hidden state of time step t-1; representing an input at a time step t; Representing a system matrix at a time step t; a selective scan output representing projection weight moment states of the infrared monitoring image features at a time step t; ; ; ; ; Wherein, the 、 Gating weight matrixes of infrared characteristics and sensor characteristics respectively; (.) Sigmoid activation function; Representing reshaping the sensor timing features into a 2D feature map format to adapt the convolution operation; conv (-) represents a 3 x3 convolution operation; (.) represents global average pooling, compressing the infrared 2D signature into a 1D sequence; And In order to gate the weight matrix, And Is a gating bias term; Representing gating as a function identifier; representing element-by-element multiplication; For preserving the proportionality coefficient of the original characteristic, the method is used for balancing the contribution of the screened characteristic and the original characteristic; Will be And Respectively and infrared intermediate quantity characteristics And gas sensor intermediate quantity feature Fusion enhancement is carried out to obtain first data and second data: ; 。
5. 5. The method for monitoring and managing multi-modal data of VOCs leaked gas based on artificial intelligence according to claim 4, wherein the first data and the second data are spliced and then subjected to linear projection to obtain fusion enhancement features : ; Wherein, the Representing a stitching operation.
6. 6. The method for multi-modal data monitoring and data management of VOCs leakage gas based on artificial intelligence according to claim 5, wherein decoding according to the fusion enhancement feature set to obtain a target scene VOCs leakage gas monitoring classification result comprises: decoding according to the fusion enhancement features at each scale in the fusion enhancement feature set in combination with a channel perception selective scanning state space decoder, wherein the decoding comprises the following steps: the fusion enhancement feature set is assumed to comprise fusion enhancement features under M scales, namely M fusion enhancement features, wherein the scales of the 1 st to M fusion enhancement features are reduced; Enhancing and upsampling the M fusion enhancement feature, restoring the scale to the scale of the M-1 fusion enhancement feature, splicing the M-1 fusion enhancement feature, inputting the spliced M-1 fusion enhancement feature into a channel perception selective scanning state space decoder for decoding, and outputting first decoding data by the channel perception selective scanning state space decoder; Up-sampling the first decoding data, restoring the scale to the scale of the M-2 fusion enhancement features, splicing the first decoding data with the M-2 fusion enhancement features, inputting the first decoding data into a channel perception selective scanning state space decoder for decoding, and outputting second decoding data by the channel perception selective scanning state space decoder; Up-sampling, feature splicing and decoding are continued according to the second decoding data until the scale of the decoding data is restored to the scale of the 1 st fusion enhancement feature, and the channel perception selective scanning state space decoder outputs decoding completion data; After decoding is completed, the fusion enhancement feature data after decoding is input into a Softmax classifier, and the Softmax classifier outputs VOCs gas leakage identification classification results of the target scene.
7. 7. An artificial intelligence based VOCs leakage gas multi-modal data monitoring and data management system for use in the method of any one of claims 1 to 6, said system comprising a first module, a second module, a third module and a fourth module; the first module is used for acquiring an infrared monitoring image of a target scene and VOCs gas sensor data; The second module is used for preprocessing the infrared monitoring image and inputting the preprocessed infrared monitoring image into the first branch for multi-scale feature extraction, wherein the feature extraction of the first branch is based on a vision selective scanning state space block, and the feature extraction of the second branch is based on a selective scanning state space block; the third module is used for fusing the characteristic data under each scale in the output of the first branch and the second branch based on the first strategy to obtain a fused enhancement characteristic set; The first strategy comprises the steps of carrying out interaction enhancement on characteristic data from a first branch and a second branch through an interaction fusion Manba model to respectively obtain first data and second data; and the fourth module is used for decoding according to the fusion enhancement feature set to obtain the VOCs gas leakage monitoring classification result of the target scene.

Description

VOCs leakage gas multi-mode data monitoring and data management method and system based on artificial intelligence Technical Field The invention relates to the technical field of computer vision, environment monitoring and artificial intelligence intersection, in particular to a VOCs leakage gas multi-mode data monitoring and data management method and system based on artificial intelligence. Background VOCs (volatile organic compounds) are widely used as pollutants in industrial production, have toxicity, flammability and strong diffusivity, and leakage of the VOCs not only can cause resource waste and destroy ecological environment, but also can cause serious safety accidents such as explosion, personnel poisoning and the like, thereby seriously threatening industrial production safety and public health. Therefore, the realization of quick, accurate and real-time monitoring of VOCs leakage is a key technical requirement for guaranteeing industrial safety production and promoting environmental protection treatment, and has irreplaceable application value in the scenes of chemical industry parks, oil gas conveying pipelines, industrial workshops and the like. The current VOCs leakage detection technology is mainly divided into two types of single-mode detection and multi-mode fusion detection, but the technology has obvious technical bottlenecks. In the single-mode detection aspect, the detection method based on the infrared image depends on the absorption characteristic imaging of VOCs to infrared light with specific wavelength, but the visual monitoring can be realized, but the problems of low resolution, fuzzy gas boundary, easiness in interference of environmental heat sources and the like of the infrared image are common, real leakage and background heat radiation artifacts are difficult to distinguish, the recognition capability of low-concentration VOCs leakage is weak, the detection method based on the VOCs gas sensor is high in response speed and sensitivity, is easy to be interfered by mixed gas, sensor data are mostly time sequence, and the long-range dependency relationship in data is difficult to effectively model based on a Convolutional Neural Network (CNN) model, so that the false alarm rate and the false alarm rate are high. Both single-mode methods have the inherent defect of single information dimension, and cannot adapt to complex environments of dynamic shielding, illumination change and coexistence of multiple interference sources in industrial scenes. Compared with the single-mode method, the prior multi-mode fusion detection technology tries to combine the space visual information of the infrared image and the time sequence concentration information of the sensor to predict the VOCs leakage, but has the key technical defects that firstly, the fusion strategy is low-efficiency, simple splicing or weighted fusion is adopted, the complementary relation of two types of data cannot be fully mined, the depth interaction of a characteristic layer cannot be realized, secondly, the method adopts a high-computation complexity transducer architecture, the fusion method is highly dependent on a self-attention mechanism, the computation amount grows in a quadratic way along with the data dimension, the high-dimensional multi-mode data is difficult to process efficiently, and the real-time detection requirement cannot be met. In recent years, SSM (STATE SPACE Model ) as an efficient sequence modeling method has gradually demonstrated great potential in the fields of natural language processing and computer vision. The SSM can effectively process global information in high-dimensional data by modeling long-range dependency relations through linear complexity. The Manba model is used as a novel architecture based on SSM, a selective scanning mechanism is introduced, model parameters can be dynamically adjusted according to input, the performance of the model in a long-sequence task is remarkably improved, and the method has high efficiency and robustness in processing high-dimensional data. However, the existing research is mainly focused on single mode, and the Manba model has not been applied to the task of detecting the VOCs leakage gas by multi-mode fusion. Therefore, a technology for monitoring VOCs leakage and managing data, which can break through the single-mode limitation, realize high-efficiency multi-mode feature fusion and adapt to complex industrial environments, is needed to solve the core technical problems of high computational complexity, insufficient long-range dependent modeling, low efficiency of multi-mode information fusion and the like in the existing method. Disclosure of Invention The invention provides a VOCs leakage gas multi-mode data monitoring and data management method and system based on artificial intelligence, which are used for solving the core technical problems of high computational complexity, insufficient long-range dependent modeling, low multi-mode informatio