Intelligent greenhouse gas identification method based on laser speckle demodulation
Through the intelligent greenhouse gas identification method based on laser speckle demodulation, the Nd:YAG laser and CNN model combined with beam expander and collimator are used to solve the problems of insufficient sensitivity and long response time of traditional greenhouse gas detection methods, and realize the Efficient and accurate gas concentration detection.
Patent Information
- Application Number
- CN202411261932.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional greenhouse gas detection methods have problems such as insufficient sensitivity, long response time and complex equipment, making it difficult to achieve sensitive, real-time and efficient detection.
Using an intelligent greenhouse gas recognition method based on laser speckle demodulation, an Nd:YAG laser is used to output 1064nm infrared laser, and speckle images are collected through a CCD camera, and a suitable CNN model is designed for image feature extraction and gas concentration prediction, combined with beam expansion The device and collimator optimize the beam propagation characteristics and use the ResNet-50 network for deep learning processing.
It realizes sensitive, real-time and efficient detection of greenhouse gas concentration, improves the accuracy and stability of the detection, and simplifies the equipment structure.
Smart Images

Figure CN118937278A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of optical gas sensing, and in particular to an intelligent greenhouse gas identification method based on laser speckle demodulation. Background Art
[0002] The concentration changes of greenhouse gases such as carbon dioxide (CO2), nitric oxide (NO), carbon monoxide (CO), etc. have an important impact on climate change. Traditional greenhouse gas detection methods (such as gas chromatography, infrared absorption spectroscopy, etc.) have problems such as insufficient sensitivity, long response time, and complex equipment. In order to solve these problems, it is crucial to provide a sensitive, real-time, and efficient detection system.
[0003] When the laser passes through an inhomogeneous medium (such as greenhouse gas in an air box), a speckle image will be formed on the CCD due to the random change of the phase relationship during the propagation of the laser. Different speckle images will be generated in different types of gases and gas media of different concentrations, so the speckle image contains the information of the gas medium.
[0004] The data is processed using a deep learning algorithm. By training a convolutional neural network (CNN) model, the system can automatically extract image features related to gas concentration and detect concentration. Summary of the invention
[0005] The purpose of the present invention is to provide an intelligent greenhouse gas identification method based on laser speckle demodulation, which is used to solve the problems of insufficient sensitivity, long response time, complex equipment, etc. existing in traditional greenhouse gas detection methods.
[0006] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions: an intelligent greenhouse gas identification method based on laser speckle demodulation, wherein the laser adopts Nd:YAG laser to output 1064nm infrared laser, and a CCD camera is used to collect speckle patterns; image preprocessing: denoising and normalizing the collected images to reduce noise interference and ensure data consistency; neural network model design and training: designing a suitable CNN model so that the CNN model can automatically extract features from the laser speckle image.
[0007] Furthermore, training data is obtained. The present invention uses Nd:YAG laser as the output laser, the transmission medium is a sealed greenhouse gas box, and CCD is used to collect speckle images. A beam expander and a collimator are placed behind the laser light source, wherein the beam expander is composed of two lenses, one is a primary lens, which is a convex lens, and the other is a secondary lens, which is a concave lens. The primary lens diverges and expands the light beam, and then the secondary lens focuses the expanded light beam again to make it a parallel beam of larger diameter; the collimator can parallelize the light beam, reduce the scattering and wavefront distortion of the light beam, and improve the stability and accuracy of the light beam when it propagates over long distances. The CCD camera is used to capture the speckle image in real time; finally, it is input into the ResNet-50 network. The deep network structure of ResNet-50 can extract more complex and abstract features, and has a good recognition ability for subtle changes and patterns in laser speckle images, and the gas concentration is predicted by processing the speckle image.
[0008] Further, data preprocessing, the speckle image collected by CCD in the above step is saved in RGB format with a pixel size of 224x224. Since the laser beam interacts with gas molecules, the collected speckle image can be used for gas identification and concentration prediction. The present invention uses the Pycharm integrated development environment and the python language to perform batch preprocessing operations on the raw data: first, the image is denoised to improve the image quality; then the pixel value range is adjusted by normalization to ensure data consistency; finally, the preprocessed image data is divided into a training set, a validation set, and a test set for subsequent model training and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Figure 1 is an experimental system diagram of an intelligent greenhouse gas identification method based on laser speckle demodulation provided by an embodiment of the present invention;
[0010] Figure 2 is a flow chart of an intelligent greenhouse gas identification method based on laser speckle demodulation provided by an embodiment of the present invention;
[0011] Figure numerals: 1 is Nd:YAG laser; 2 is beam expander; 3 is collimator; 4 is greenhouse gas box; 5 is gas cylinder one; 6 is gas cylinder two; 7 is CCD camera; 8 is deep learning network model. DETAILED DESCRIPTION
[0012] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The steps in the following embodiments are only for the convenience of explanation and description, and the order between the steps is not limited. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.
[0013] See also Figure 1 The embodiment of the present invention provides an experimental system diagram of an intelligent greenhouse gas identification method based on laser speckle demodulation, including an Nd:YAG laser 1, a beam expander 2, a collimator 3, a greenhouse gas box 4, a gas cylinder 1 5, a gas cylinder 2 6, a CCD camera 7, and a deep learning network model 8; the Nd:YAG laser light source first emits a laser, and the beam is expanded and shaped by the beam expander and the collimator; then the laser enters the greenhouse gas box, and the gas box is connected to two gas cylinders, and the greenhouse gas concentration in the gas box can be adjusted by the gas cylinders; in the gas box, the laser interacts with the gas molecules; a CCD camera is used behind the gas box to collect and form a speckle image carrying information on the type and concentration of the gas; finally, a convolutional neural network (CNN) is used to process and analyze the collected speckle image to predict the type and concentration of the gas.
[0014] Specifically, the Nd:YAG laser emits a near-infrared laser with a wavelength of 1064nm. The laser diameter is expanded four times by the beam expander to ensure that the laser beam has a larger contact volume with the gas in the gas box, so as to more comprehensively detect the distribution of gas concentration. At the same time, the beam expander can reduce the divergence angle of the beam and reduce the impact of the beam scattering and absorption in the gas, thereby improving the stability and accuracy of the measurement. Then, through the collimator, the collimator can further organize the beam after expansion into a parallel beam, reduce the divergence and scattering of the beam, and thus improve the accuracy of the measurement. By using the collimator in conjunction with the beam expander, the propagation characteristics of the beam can be optimized, the performance of the gas concentration measurement system can be improved, and the measurement accuracy can be improved.
[0015] Furthermore, the light beam optimized by the collimator and beam expander is injected into the greenhouse gas box to interact with the gas molecules therein. The greenhouse gas box is a fully enclosed transparent rectangular box with a size of 20cm*20cm*100cm. The box is connected to two gas cylinders. The corresponding gas cylinders are selected according to the type of gas to be predicted (such as carbon dioxide, methane, nitrogen oxides, etc.). In this experiment, we chose carbon dioxide gas cylinders and methane gas cylinders. The gas cylinders are equipped with flow meters, which are used to accurately control the flow of gas, so as to adjust the concentration of gas in real time. The greenhouse gas box provides a stable and controllable environment for the interaction between laser and gas molecules, reducing external interference. When the laser beam passes through the gas in the box, the gas molecules will interact with the laser beam, including light scattering and absorption, and the light scattering and absorption will change the intensity and speckle pattern of the beam. Due to the different types and concentrations of gases, the changes caused are also different. The speckle image collected by CCD contains the type and concentration information of the gas, and the type and concentration of the gas can be predicted by demodulating the speckle image.
[0016] See also Figure 2, an embodiment of the present invention provides a flow chart of an intelligent greenhouse gas identification method based on laser speckle demodulation. First, the speckle image is collected, and then the collected image is preprocessed, mainly including the preprocessing operations of scaling, denoising, and normalization of the speckle image, and then the model is trained. The preprocessed data set is first divided into a training set, a validation set, and a test set. The model adopts the ResNet-50 pre-trained model, and the model is trained using the training set; then the new speckle image is input into the trained ResNet-50 model to predict the gas type and concentration; finally, the accuracy of the model is evaluated by comparing with the actual results.
[0017] Specifically, the preprocessed image is scaled to 224×224 pixels, which is the input size required by the ResNet-50 network. The processed image can be input into the ResNet-50 model, first passing through the convolutional layer and residual block for feature extraction; then entering the global average pooling and fully connected layer; and finally outputting a scalar, which is the prediction of gas concentration.
[0018] Furthermore, the ResNet-50 network used in this system has 49 convolutional layers and 16 residual blocks. The convolutional layers are used to extract the edges and textures of the images. The residual blocks contain skip connections, which allow the network to transmit information across layers and extract high-level features. At the same time, each convolutional layer contains a ReLU activation function, which can increase the nonlinearity of the data and enable the model to learn complex feature patterns. Then it enters the global average pooling layer. The input of the global average pooling layer is the output feature map of the last residual block (Conv5_x) with a size of 7x7x2048 (7x7 represents the spatial dimension of the feature map, and 2048 represents the number of channels of the feature map). The obtained 2048-dimensional feature vector is input into the fully connected layer, and the input 2048-dimensional feature vector is mapped to the final output through the weight matrix. The dimension of the weight matrix is 1x2048, that is, 2048 input features are mapped to one output value. The fully connected layer finally outputs a scalar value, which is the predicted gas concentration.
[0019] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. An intelligent greenhouse gas identification method based on laser speckle demodulation belongs to the field of optical gas sensing, and its characteristics are: Nd: YAG laser (1), beam expander (2), collimator (3), greenhouse gas box (4), gas cylinder 1 (5), gas cylinder 2 (6), CCD camera (7), deep learning network model (8), the specific steps of the intelligent greenhouse gas identification method based on laser speckle demodulation are as follows: 1) The laser uses a Nd:YAG laser (1) to output 1064nm infrared laser; 2) Connecting a beam expander (2) and a collimator (3) to an Nd:YAG laser (1), the laser passes through the beam expander to expand the laser diameter by four times, and the beam expander can reduce the divergence angle of the light beam, reduce the influence of the light beam scattering and absorption in the gas, thereby improving the stability and accuracy of the measurement; and then passing through the collimator, the collimator can further organize the expanded light beam into a parallel light beam, reduce the divergence and scattering of the light beam, thereby improving the accuracy of the measurement; 3) The light beam optimized by the collimator and beam expander is injected into the greenhouse gas box (4) to interact with the gas molecules therein; the concentration and type of gas in the greenhouse gas box are determined by two sealed gas cylinders; 4) The speckle image of the laser beam after interacting with greenhouse gases is captured by a CCD camera (7), and then input into a deep learning network model (8) for information demodulation.
2. The intelligent greenhouse gas identification method based on laser speckle demodulation according to claim 1 is characterized in that: The beam expander is composed of two lenses, one is the primary lens, which is a convex lens, and the other is the secondary lens, which is a concave lens; the primary lens diverges and expands the light beam, and then the secondary lens focuses the expanded light beam again to make it a parallel beam with a larger diameter; the collimator can parallelize the light beam, reduce the scattering and wavefront distortion of the light beam, and improve the stability and accuracy of the light beam when it propagates over long distances.
3. The intelligent greenhouse gas identification method based on laser speckle demodulation according to claim 1 is characterized in that: The greenhouse gas box is a sealed box with a size of 20cm*20cm*100cm. The box is connected to two gas cylinders with flow meters on them. The flow meters are used to accurately control the flow of the gas, thereby adjusting the concentration of the gas in real time. The greenhouse gas box provides a stable and controllable environment for the interaction between the laser and the gas molecules, reducing external interference.
4. The intelligent greenhouse gas identification method based on laser speckle demodulation according to claim 1 is characterized in that: The ResNet-50 network is used. The deep network structure of ResNet50 can extract more complex and abstract features, and has better recognition ability for subtle changes and patterns in laser speckle images.
