Intelligent environment-friendly sensing method and device for spraying operation environment

Abstract

The invention provides an intelligent environment-friendly sensing method and device for a spraying operation environment, and the method comprises the steps: carrying out the multi-scale feature fusion through employing a convolutional neural network in combination with a feature pyramid network, and obtaining an environment visibility index EVI of the spraying operation environment; based on the Mie scattering principle, a dust concentration inversion model is established, and the dust concentration PM10 is obtained; and the environment visibility index EVI and the dust concentration PM10 are fused to obtain the air quality index AQI. Through fusion of artificial intelligence image recognition and advanced optical sensing technology and assistance of an innovative data fusion and exception filtering mechanism, the environment visibility and the dust particulate matter quantity of a spraying operation environment can be intelligently monitored, the safety of the operation environment is ensured, the air supply amount and the air exhaust amount of a coating workshop can be accurately controlled, and the working efficiency is improved. And the sight safety of operators is guaranteed while energy is saved, potential safety hazards are reduced, the device can adapt to the complex spraying environment, high practicability and reliability are achieved, and enterprises are helped to meet the requirements of environmental protection laws and regulations.

Description

Technical Field

[0001] The present invention relates to an intelligent environmentally friendly sensing method and device for a spraying operation environment. Background Art

[0002] Spray painting plants are specialized buildings used for spray painting operations. During the spraying process, large quantities of particulate matter are generated. Inadequate particulate matter detection in the spray painting plant makes it difficult to monitor dust and paint mist concentrations in real time, making it difficult to detect quality issues such as uneven spraying, missed spraying, and overspraying. Furthermore, excessive concentrations of dust and harmful gases pose a threat to worker health and violate environmental regulations. Reduced visibility can affect operators' vision and increase safety risks.

[0003] Existing particulate matter detection devices have limited functionality and a limited measurement range. While they offer high accuracy for detecting small atmospheric particles, they are ineffective for detecting large particles in coating production operations. Even the same product from the same manufacturer can produce widely varying results under the same conditions, failing to meet the evolving demands for comprehensive and real-time particle detection in spray painting environments. Summary of the Invention

[0004] The present invention provides an intelligent environmental protection sensing device for a spraying operation environment, which monitors the environmental visibility and the amount of dust particles in the spraying operation environment in real time to ensure the safety of the operation environment.

[0005] The present invention solves the above technical problems through the following technical solutions:

[0006] The present invention provides an intelligent environmental protection perception method for a spraying operation environment, comprising:

[0007] S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment;

[0008] S200 , using a convolutional neural network in combination with a feature pyramid network to perform multi-scale feature fusion on the environmental image information to obtain an environmental visibility index (EVI) of the spraying operation environment.

[0009] In this technical solution, the powerful feature extraction capability of convolutional neural networks (CNN) is utilized, combined with the multi-scale feature fusion advantage of feature pyramid networks (FPN), to learn from environmental image information and quantify the degree of environmental blur (visibility) through the environmental visibility index (EVI).

[0010] Preferably, the intelligent environmental protection perception method further includes:

[0011] S300: Using a dust sensor to collect scattered light signals from the spraying operation environment, establishing a dust concentration inversion model based on the Mie scattering principle, and obtaining the dust concentration PM10.

[0012] In this technical solution, based on the Mie scattering theory, the difference in dual-wavelength laser scattering is used to invert the particle size distribution, and then the dust concentration PM10 is obtained; the environment is comprehensively evaluated by the environmental visibility index EVI and the dust concentration PM10, which is more objective and accurate than using only the environmental visibility index EVI or the dust concentration PM10.

[0013] Preferably, the intelligent environmental protection perception method further includes:

[0014] S400: Obtain the air quality index AQI according to the environmental visibility index EVI and the dust concentration PM10, AQI=W EVI ×(1-EVI)×100+(1-W EVI )×PM10 / 2,W EVI is the environmental visibility weight coefficient, 0.3≤W EVI ≤0.94; PM10<100, W EVI Take the value as 0.8; when PM10<500, W EVI Take the value as 0.5; when PM10>500, W EVI The value is 0.3.

[0015] In this technical solution, the air quality index (AQI) obtained through the above calculation is a comprehensive indicator that comprehensively reflects the impact of environmental visibility deterioration (caused by particulate matter such as dust) and the dust concentration itself on air quality. It overcomes the limitations of relying solely on the environmental visibility index (EVI) or dust concentration PM10, and provides managers with a more comprehensive environmental status assessment.

[0016] Preferably, the step S200 includes:

[0017] Step S201: pre-process the environmental image information to obtain an input image, where the input image is a 512×512 pixel RGB image;

[0018] Step S202: The convolutional neural network uses ResNet-50 as the backbone network, inputs the input image into the convolutional neural network, and extracts three feature maps Stage 3, Stage 4, and Stage 5. The resolution of Stage 3 is 64×64×24, the resolution of Stage 4 is 32×32×40, and the resolution of Stage 5 is 16×16×80.

[0019] Step S203: using the feature pyramid network and a top-down path, upsampling the Stage 5 by a factor of 2, and fusing the Stage 5, Stage 4, and Stage 3 step by step by element-by-element addition;

[0020] Step S204: Use 1×1 convolution to adjust the channel lateral connection to generate multi-scale feature maps P3, P4 and P5, where the resolution of P3 is 64×64, the resolution of P4 is 32×32, and the resolution of P5 is 16×16;

[0021] Step S205: Process the feature maps P3, P4, and P5 using formula (1) respectively.

[0022] Ffusedi=Conv1×1(Pi)+α·UpSample(Ffusedi+1) (1)

[0023] Among them, Ffused is the fused feature map, Conv1X1 is used to adjust the number of channels, α is the fusion weight coefficient, and i is 3, 4, or 5;

[0024] Step S206: perform global average pooling on the fused feature map F_fused, output a 512-dimensional feature vector, and map it to the environmental visibility index EVI using formula (2) through the fully connected layer;

[0025] EVI=σ(w T ·GAP(Ffused)+b) (2)

[0026] Among them, σ is the Sigmoid activation function (output normalized to 0-1), w is the weight vector, GAP is the global average pooling of the image, Ffused is the multi-scale fused feature image, b is the bias term, and w and b are trainable parameters.

[0027] In this technical solution, ResNet-50 is used as the backbone network, combined with the Feature Pyramid Network (FPN) to obtain the EVI value mapped to the [0,1] interval; the closer the EVI value is to 1, the better the visibility; the closer it is to 0, the worse the visibility.

[0028] Preferably, the step S200 further includes:

[0029] Step S207: Establish dust concentration 0-2000 μg / m 3A manually calibrated dataset of environmental image information is prepared, wherein the artificial visibility is graded into artificial visibility level 1, artificial visibility level 2, and artificial visibility level 3. Artificial visibility level 1 is "good visibility, with clear equipment outlines", artificial visibility level 2 is "medium visibility, with slight blur", and artificial visibility level 3 is "poor visibility, with no ability to identify key equipment";

[0030] Step S208: Training the convolutional neural network and the feature pyramid network through the mean square error loss function so that the EVI matches the artificial visibility level. The EVI ≥ 0.8 corresponds to the first level of the artificial visibility level, the 0.5 ≤ EVI < 0.8 corresponds to the second level of the artificial visibility level, and the EVI < 0.5 corresponds to the third level of the artificial visibility level.

[0031] In this technical solution, through training, the EVI output value of the model is made to correspond to the manual grading standard, which makes EVI no longer an abstract score, but a quantitative indicator closely related to human subjective perception and operational safety requirements.

[0032] Preferably, step S300 includes:

[0033] S301, the dust sensor is a dual-wavelength laser scattering sensor array, which emits a dual-wavelength laser beam into the spraying operation environment to form a scattered light signal of dust particles, and the main wavelength of the dual-wavelength laser scattering sensor array is 650nm and the auxiliary wavelength is 850nm;

[0034] S302, receiving light signals scattered by dust particles through a signal conditioning circuit, and separating the scattered light of the 650nm channel and the 850nm channel respectively;

[0035] S303. Convert the optical signal into an electrical signal through the 650nm channel and the 850nm channel, wherein the 650nm channel uses a transimpedance amplifier with a gain of 1 MΩ, and the 850nm channel uses a transimpedance amplifier with a gain of 500 kΩ. The bandwidth of the 650nm channel is 10 kHz, and the bandwidth of the 850nm channel is 10 kHz.

[0036] S304, performing ADC sampling on the electrical signal with a sampling resolution of 16 bits and a sampling rate of 10 kSPS to generate a dual-wavelength digital signal;

[0037] S305. Establish a single particle scattering intensity model based on Mie scattering theory, as shown in formula (3):

[0038]

[0039] Where I is the scattered light intensity, S is the Mie scattering complex amplitude function, which is numerically solved by the Bessel function; λ is the specific wavelength; θ is the scattering angle, d is the particle size, and m is the complex refractive index;

[0040] S306, using a pre-built dual-wavelength lookup table LUT, performing particle size inversion on the digital signal by non-negative least squares method, wherein the LUT is generated by pre-calculating the scattering intensity of different d and m combinations at wavelengths of 650nm and 850nm,

[0041] Among them, the particle size range d is 0.3-10 μm, the step size λ is 0.1 μm, and the refractive index m∈[1.3+0i,1.7+0.1i];

[0042] S307. Output the dust concentration PM10 according to the inversion result.

[0043] In this technical solution, based on the Mie Scattering theory, the particle size distribution of particulate matter is inverted using the difference in dual-wavelength laser scattering, and then the dust concentration PM10 is calculated, which can significantly improve the detection accuracy of the dust concentration PM10.

[0044] Preferably, the intelligent environmental perception method further includes step S350, which is before step S400 and after step S300. Step S350 uses a deep learning algorithm to detect and filter instantaneous abnormal values ​​in the environmental visibility index EVI and dust concentration PM10.

[0045] In this technical solution, the detected abnormal points are filtered, which can effectively filter out short-term pulse-type interference (spikes) while better retaining the true trend of the data.

[0046] Preferably, the step S350 includes:

[0047] Step S351: Build an environmental visibility database, collect a number of spray painting workshop scene images with different lighting conditions, dust concentrations, and equipment layouts, and annotate each of the spray painting workshop scene images with a manually scored environmental visibility index (EVI);

[0048] Step S352: performing anti-interference data enhancement on the spray painting workshop scene image to form a simulated interference image;

[0049] Step S353: Build a deep learning model that simultaneously performs EVI regression prediction and key area segmentation, and design a multi-task loss function L = λ1·L_EVI+λ2·L_seg;

[0050] Where L_EVI is the mean square error loss of EVI regression, L_seg is the Dice loss of key region segmentation, which is calculated as 1-2×(predicted segmentation area ∩ true segmentation area) / (predicted area + true area); set the weight coefficients λ1=1.0, λ2=0.5;

[0051] Using the simulated interference image to train the deep learning model until convergence;

[0052] Step S354: Input the environmental image information into the trained deep learning model, and synchronously output the EVI prediction value and the dust area segmentation map associated with PM10; based on the time sliding window statistics of the instantaneous fluctuations of EVI and PM10, identify abnormal values ​​exceeding the preset threshold; filter the instantaneous abnormality detection results through the median filtering algorithm.

[0053] In this technical solution, by constructing a model to simultaneously perform two tasks of EVI regression prediction and key area segmentation, combined with the inconsistency analysis of the model output (for EVI / image anomalies) and statistical anomaly detection of time series (for PM10 and EVI), instantaneous anomalies can be identified, and short-term pulse-type interference (spikes) can be effectively filtered out, while better retaining the true trend of the data.

[0054] The present invention also provides an intelligent environmental protection perception device for a spraying operation environment, which uses the above-mentioned intelligent environmental protection perception method. The intelligent environmental protection perception device includes an image acquisition module, an image processing module, a dust analysis module and a data fusion module. The image acquisition module is used to collect environmental image information of the spraying operation environment; the image processing module is used to use a convolutional neural network to combine the feature pyramid network to perform multi-scale feature fusion on the environmental image information to obtain the environmental visibility index EVI of the spraying operation environment; the dust analysis module is used to collect the scattered light signal of the spraying operation environment, and based on the Mie scattering principle, establish a dust concentration inversion model to obtain the dust concentration PM10; the data fusion module is used to receive the environmental visibility index EVI of the image processing module and the dust concentration PM10 of the dust analysis module, and fuse the environmental visibility index EVI and the dust concentration PM10 to obtain the air quality index AQI.

[0055] In this technical solution, the air quality index AQI is obtained through the setting of the above-mentioned device, which overcomes the limitations of relying solely on the environmental visibility index EVI or the dust concentration PM10, and provides managers with a more comprehensive environmental status assessment.

[0056] Preferably, the intelligent environmental perception device also includes a data filtering module, which is arranged between the image processing module, the dust analysis module and the data fusion module. The data filtering module is used to receive the environmental visibility index EVI of the image processing module and the dust concentration PM10 of the dust analysis module. The data filtering module is also used to use a deep learning algorithm to detect and filter instantaneous abnormal values ​​in the environmental visibility index EVI and dust concentration PM10.

[0057] In this technical solution, by combining the inconsistency analysis of the model output (for EVI / image anomalies) and the statistical anomaly detection of the time series (for PM10 and EVI), instantaneous anomalies can be identified, and short-term pulse-type interference (spikes) can be effectively filtered out, while better retaining the true trend of the data.

[0058] On the basis of conforming to the common sense in this field, the above-mentioned preferred conditions can be arbitrarily combined to obtain the preferred embodiments of the present invention.

[0059] The positive progress effect of the present invention is:

[0060] The above-mentioned intelligent environmental protection perception method and device for the spraying operation environment, by integrating artificial intelligence image recognition and advanced optical sensing technology, and supplemented by innovative data fusion and abnormal filtering mechanisms, can intelligently monitor the environmental visibility and the number of dust particles in the spraying operation environment to ensure the safety of the working environment. It can accurately control the air supply and exhaust volume of the painting workshop, while saving energy. It ensures the safety of the operator's sight and reduces safety hazards. It can adapt to complex spraying environments, has strong practicality and reliability, and helps enterprises meet environmental protection regulations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Shown is a schematic diagram of the steps of Example 1 of the intelligent environmentally friendly perception method for the spraying operation environment of the present invention.

[0062] Figure 2 Shown Figure 1 The diagram shown is a schematic diagram of steps 201 to 206 of step 200 of the intelligent environmental protection perception method for the spraying operation environment.

[0063] Figure 3 Shown Figure 1 The diagram shown is a schematic diagram of steps 207 to 208 of step 200 of the intelligent environmental protection perception method for the spraying operation environment.

[0064] Figure 4 Shown is a schematic diagram of the steps of Example 2 of the intelligent environmentally friendly perception method for the spraying operation environment of the present invention.

[0065] Figure 5 Shown Figure 4The diagram shown is a schematic diagram of step 300 of the method for intelligent environmental protection perception of the spraying operation environment.

[0066] Figure 6 Shown is a schematic diagram of the steps of Example 3 of the intelligent environmental protection perception method for the spraying operation environment of the present invention.

[0067] Figure 7 Shown Figure 6 The schematic diagram of the structure of the device corresponding to the intelligent environmental protection perception method of the spraying operation environment is shown.

[0068] Figure 8 Shown is a schematic diagram of the steps of Example 4 of the intelligent environmentally friendly perception method for the spraying operation environment of the present invention.

[0069] Figure 9 Shown Figure 8 FIG. 3 is a schematic diagram of step 350 of the method for intelligent environmentally friendly perception of a spraying operation environment.

[0070] Figure 10 Shown Figure 8 The schematic diagram of the structure of the device corresponding to the intelligent environmental protection perception method of the spraying operation environment is shown.

[0071] Description of Reference Numerals

[0072] Image acquisition module 1

[0073] Image processing module 2

[0074] Dust analysis module 3

[0075] Data fusion module 4

[0076] Data storage module 5

[0077] Data filtering module 6 DETAILED DESCRIPTION

[0078] The present invention is further described below by way of examples, but the present invention is not limited to the scope of the examples.

[0079] Example 1

[0080] like Figure 1 As shown, this embodiment provides an intelligent environmentally friendly perception method for a spraying operation environment, including:

[0081] S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment;

[0082] S200, a convolutional neural network is used to perform multi-scale feature fusion on the environmental image information in combination with a feature pyramid network to obtain the environmental visibility index (EVI) of the spraying operation environment.

[0083] By utilizing the powerful feature extraction capability of convolutional neural networks (CNN) and combining the multi-scale feature fusion advantages of feature pyramid networks (FPN), we can learn from environmental image information and quantify the degree of environmental blur (visibility) through the environmental visibility index (EVI).

[0084] like Figure 2 Specifically, step S200 includes:

[0085] Step S201: Preprocess the environmental image information to obtain an input image, which is a 512×512 pixel RGB image. The RGB image can also be enhanced by fog filtering.

[0086] Step S202: The convolutional neural network uses ResNet-50 as the backbone network, inputs the input image into the convolutional neural network, and extracts three feature maps Stage 3, Stage 4, and Stage 5. The resolution of Stage 3 is 64×64×24, the resolution of Stage 4 is 32×32×40, and the resolution of Stage 5 is 16×16×80.

[0087] Step S203: Use a feature pyramid network and a top-down path to upsample Stage 5 by a factor of 2, and fuse Stage 5, Stage 4, and Stage 3 step by step by element-by-element addition.

[0088] Step S204: Use 1×1 convolution to adjust the channel lateral connection to generate multi-scale feature maps P3, P4 and P5. The resolution of P3 is 64×64, the resolution of P4 is 32×32, and the resolution of P5 is 16×16.

[0089] Specifically, the specific operations for generating multi-scale feature maps P3, P4, and P5 are shown in Table 1.

[0090] Table 1. Steps for generating multi-scale feature maps P3, P4 and P5

[0091] Hierarchy Input Source operate Output size P3 Stage 3 (64×64×24) 1×1 convolution adjusts the number of channels 64×64×64 P4 Stage 4 (32×32×40) 3×3 convolution + 2x upsampling 64×64×64 P5 Stage 5 (16×16×80) 3×3 convolution + 4x upsampling 64×64×64

[0092] The fusion formula used is as follows:

[0093] Pout=Conv3×3(UpSample(Phigher)+Conv1×1(Pcurrent) (4)

[0094] Step S205: Process the feature maps P3, P4, and P5 using formula (1) respectively.

[0095] Ffusedi=Conv1×1(Pi)+α·UpSample(Ffusedi+1) (1)

[0096] Among them, Ffused is the fused feature map, Conv1X1 is used to adjust the number of channels, α is the fusion weight coefficient, and i is 3, 4, or 5;

[0097] Step S206: perform global average pooling on the fused feature map F_fused, output a 512-dimensional feature vector, and map it to the environmental visibility index EVI using formula (2) through the fully connected layer;

[0098] EVI=σ(w T ·GAP(Ffused)+b) (2)

[0099] Among them, σ is the Sigmoid activation function (output normalized to 0-1), w is the weight vector, GAP is the global average pooling of the image, Ffused is the multi-scale fused feature image, b is the bias term, and w and b are trainable parameters.

[0100] The above method uses ResNet-50 as the backbone network to process the input 512x512 pixel RGB image. At different stages of the deep network (such as Stage 3, 4, and 5), feature maps with different spatial resolutions and semantic levels are extracted. For example, the low-level feature maps (Stage 3) have high resolution and contain rich details (such as edges and textures); the high-level feature maps (Stage 5) have low resolution and contain stronger semantic information (such as object category and overall structure) but are less sensitive to details.

[0101] Reduced visibility (blurring) in a spray painting environment can be caused by diffuse dust (global impact) or localized high concentration areas (local impact), and has multi-scale characteristics. The Feature Pyramid Network (FPN) constructs a feature pyramid (P3, P4, P5) that fuses high-, mid-, and low-level information through a top-down path (upsampling high-level features) and lateral connections (fusing low-level features of the same resolution). Step S205 further strengthens the fusion and information transfer between feature maps of different scales, enabling the model to simultaneously capture global turbidity and the blurring of key local details.

[0102] Finally, global average pooling (GAP) is performed on the fused feature map, compressing it into a fixed-length feature vector (e.g., 512 dimensions) that highly summarizes the overall visibility information of the image. Finally, a fully connected layer maps the feature vector to an EVI value in the range [0, 1]. EVI values ​​closer to 1 indicate better visibility, while values ​​closer to 0 indicate worse visibility.

[0103] like Figure 3 As shown, based on steps S201 to S206, step S200 may further include:

[0104] Step S207: Establish dust concentration 0-2000 μg / m 3 The artificial visibility level is divided into artificial visibility level 1, artificial visibility level 2, and artificial visibility level 3. Artificial visibility level 1 is "good visibility, clear equipment outline", artificial visibility level 2 is "medium visibility, slightly blurred", and artificial visibility level 3 is "poor visibility, unable to identify key equipment";

[0105] Step S208: Train the convolutional neural network and the feature pyramid network using the mean square error loss function to make the EVI match the artificial visibility level. EVI ≥ 0.8 corresponds to the first level of artificial visibility, 0.5 ≤ EVI < 0.8 corresponds to the second level of artificial visibility, and EVI < 0.5 corresponds to the third level of artificial visibility.

[0106] By establishing a typical dust concentration range (0~2000μg / m 3 ) image dataset, and manually graded and labeled them strictly according to the clarity standard (artificial visibility level 1 is "good visibility, clear equipment outline", artificial visibility level 2 is "medium visibility, slightly blurred", and artificial visibility level 3 is "poor visibility, key equipment cannot be identified"). When training the convolutional neural network (CNN) + feature pyramid network (FPN) model, the mean square error loss function (MSE) is used for supervised learning, forcing the model's EVI output value to correspond to the manual grading standard (EVI ≥ 0.8 corresponds to level 1, 0.5 ≤ EVI < 0.8 corresponds to level 2, and EVI < 0.5 corresponds to level 3). This makes EVI no longer an abstract score, but a quantitative indicator closely related to human subjective perception and operational safety requirements.

[0107] Example 2

[0108] This embodiment further includes step 300 based on the first embodiment.

[0109] like Figure 4 As shown, this embodiment provides an intelligent environmentally friendly perception method for a spraying operation environment, including:

[0110] S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment;

[0111] S200: Use convolutional neural network and feature pyramid network to perform multi-scale feature fusion on environmental image information to obtain the environmental visibility index (EVI) of the spraying operation environment;

[0112] S300: Use a dust sensor to collect scattered light signals from the spraying environment. Based on the Mie scattering principle, establish a dust concentration inversion model to obtain the dust concentration PM10.

[0113] Based on Mie scattering theory, the system uses the difference in dual-wavelength laser scattering to invert the particle size distribution, thereby obtaining the PM10 concentration. This combined evaluation of the environmental visibility index (EVI) and PM10 is more objective and accurate than using either index alone.

[0114] like Figure 5 Specifically, step S300 includes:

[0115] S301, the dust sensor is a dual-wavelength laser scattering sensor array, which emits a dual-wavelength laser beam into the spraying operation environment to form a scattered light signal of dust particles. The main wavelength of the dual-wavelength laser scattering sensor array is 650nm and the auxiliary wavelength is 850nm;

[0116] S302, receiving light signals scattered by dust particles through a signal conditioning circuit, and separating the scattered light of the 650nm channel and the 850nm channel respectively;

[0117] S303 converts optical signals into electrical signals through the 650nm channel and the 850nm channel. The 650nm channel uses a transimpedance amplifier with a gain of 1MΩ, and the 850nm channel uses a transimpedance amplifier with a gain of 500kΩ. The bandwidth of the 650nm channel is 10kHz, and the bandwidth of the 850nm channel is 10kHz. The 650nm red light can sensitively capture fine particles of 0.3-2.5μm; and the 850nm red light can efficiently detect coarse particles of 2.5-10μm, solving the problem of single-wavelength sensors missing mixed dust detection.

[0118] S304, performing ADC sampling on the electrical signal with a sampling resolution of 16 bits and a sampling rate of 10 kSPS to generate a dual-wavelength digital signal;

[0119] S305. Establish a single particle scattering intensity model based on Mie scattering theory, as shown in formula (3):

[0120]

[0121] Where I is the scattered light intensity, S is the Mie scattering complex amplitude function, which is numerically solved by the Bessel function; λ is the specific wavelength; θ is the scattering angle, d is the particle size, and m is the complex refractive index;

[0122] S306. Using a pre-built dual-wavelength lookup table (LUT), perform particle size inversion on the digital signal using the non-negative least squares method. The LUT is generated by pre-calculating the scattering intensity of different d and m combinations at wavelengths of 650nm and 850nm. The LUT method supports adapting to different dust types by updating the database.

[0123] Among them, the particle size range d is 0.3-10 μm, the step size λ is 0.1 μm, and the refractive index m∈[1.3+0i,1.7+0.1i];

[0124] S307. Output the dust concentration PM10 according to the inversion result.

[0125] Spray dust particles have a wide size distribution (ranging from submicron paint mist to larger particles), and the sensitivity of single-wavelength scattering to particle size varies across different size ranges and is easily affected by the uncertainty of the particle's refractive index. A dual-wavelength system employing a primary wavelength of 650nm (red light, more sensitive to smaller particles) and a secondary wavelength of 850nm (near-infrared, more sensitive to larger particles) significantly improves the accuracy and robustness of particle size distribution inversion by leveraging the difference in scattering signals at these two wavelengths. This system is particularly effective in detecting large particles (>1μm), which are common in spray environments, compared to single-wavelength sensors.

[0126] The scattered light generated by laser irradiation of particulate matter in the air is separated and received through precise optical design and signal conditioning circuitry. The scattered light signals of the 650nm and 850nm channels are then received. A transimpedance amplifier is then used to convert the weak photocurrent into a voltage signal. Different gains (1MΩ vs. 500kΩ) take into account the possible signal strength differences between the two channels, while a high-precision ADC ensures the digitization quality of the signal.

[0127] Formula (3) describes the scattered light intensity I of a single spherical particle at a specific wavelength λ, scattering angle θ, particle size d, and complex refractive index m. Here, S(λ,θ,d,m) is a complex Mie scattering amplitude function that must be solved using numerical methods (such as Bessel functions).

[0128] Then, based on the characteristics of particulate matter in the spraying environment, a dual-wavelength scattering intensity lookup table (LUT) covering the expected particle size range (d: 0.3-10μm, step size 0.1μm) and the possible complex refractive index range (m∈[1.3+0i,1.7+0.1i]) is pre-calculated; and the digital signals of the 650nm and 850nm channels actually obtained are optimally fitted with the simulated signals under different particle size distribution assumptions in the LUT using the non-negative least squares method (NNLS). The non-negative least squares method (NNLS) can effectively solve the particle size distribution (PSD) and ensure that the result is non-negative (physical meaning requirement). Finally, based on the inverted particle size distribution, the mass concentration of particles with an aerodynamic diameter ≤10μm, i.e. PM10, is calculated.

[0129] Example 3

[0130] This embodiment further includes step 400 based on the second embodiment.

[0131] like Figure 6 As shown, this embodiment provides an intelligent environmentally friendly perception method for a spraying operation environment, including:

[0132] S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment;

[0133] S200: Use convolutional neural network and feature pyramid network to perform multi-scale feature fusion on environmental image information to obtain the environmental visibility index (EVI) of the spraying operation environment;

[0134] S300: Use a dust sensor to collect scattered light signals from the spraying environment, establish a dust concentration inversion model based on the Mie scattering principle, and obtain the dust concentration PM10;

[0135] S400: Obtain the air quality index AQI based on the environmental visibility index EVI and the dust concentration PM10, AQI = W EVI ×(1-EVI)×100+(1-WEVI)×PM10 / 2, W EVI is the environmental visibility weight coefficient, 0.3≤W EVI ≤0.94.

[0136] Among them, the environmental visibility weight coefficient W EVI , which determines the relative importance of ambient visibility index EVI and dust concentration PM10 in the final AQI.

[0137] W EVI It is not a fixed value. Its setting needs to consider the focus of specific application scenarios. EVIIt is necessary to dynamically match the current operating environment changes to achieve integration. When PM10<100, W EVI Take the value as 0.8; PM10<500, then W EVI Take the value as 0.5; PM10>500, then W EVI The value is 0.3;

[0138] The Air Quality Index (AQI) calculated above is a comprehensive indicator that reflects the impact of both deteriorating visibility (caused by particulate matter like dust) and dust concentration itself on air quality. Higher values ​​indicate poorer air quality and higher environmental risks. The AQI overcomes the limitations of relying solely on the Environmental Visibility Index (EVI) or PM10 dust concentration, providing managers with a more comprehensive assessment of environmental conditions.

[0139] like Figure 7 As shown, the intelligent environmental protection perception device for the spraying operation environment corresponding to the above method includes an image acquisition module 1, an image processing module 2, a dust analysis module 3 and a data fusion module 4. The image acquisition module 1 is used to collect environmental image information of the spraying operation environment; the image processing module 2 is used to use a convolutional neural network to perform multi-scale feature fusion on the environmental image information in combination with a feature pyramid network to obtain the environmental visibility index EVI of the spraying operation environment; the dust analysis module 3 is used to collect the scattered light signal of the spraying operation environment, and based on the Mie scattering principle, establish a dust concentration inversion model to obtain the dust concentration PM10; the data fusion module 4 is used to receive the environmental visibility index EVI of the image processing module and the dust concentration PM10 of the dust analysis module, and fuse the environmental visibility index EVI and the dust concentration PM10 to obtain the air quality index AQI.

[0140] The above-mentioned environmental visibility index EVI, dust concentration PM10 and air quality index AQI can be stored using the data storage module 5 and output to an external display device. The data display method can be displayed in the form of an environmental quality report. The display content of the environmental quality report is shown in Table 2.

[0141] Table 2 Display contents of environmental quality report

[0142]

[0143] Example 4

[0144] This embodiment further includes step S350 based on the third embodiment.

[0145] like Figure 8 As shown, this embodiment provides an intelligent environmentally friendly perception method for a spraying operation environment, including:

[0146] S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment;

[0147] S200: Use convolutional neural network and feature pyramid network to perform multi-scale feature fusion on environmental image information to obtain the environmental visibility index (EVI) of the spraying operation environment;

[0148] S300: Use a dust sensor to collect scattered light signals from the spraying environment, establish a dust concentration inversion model based on the Mie scattering principle, and obtain the dust concentration PM10;

[0149] S350 uses deep learning algorithms to detect and filter out instantaneous abnormal values ​​in the environmental visibility index (EVI) and dust concentration (PM10);

[0150] S400: Obtain the air quality index AQI based on the environmental visibility index EVI and the dust concentration PM10, AQI = W EVI ×(1-EVI)×100+(1-W EVI )×PM10 / 2,W EVI is the environmental visibility weight coefficient, 0.3≤W EVI ≤0.94; when PM10<100, W EVI Take the value as 0.8; when PM10<500, W EVI Take the value as 0.5; when PM10>500, W EVI The value is 0.3.

[0151] Image acquisition can be disrupted by transient strong light (such as from welding), lens obstructions, and splashing objects. Dust sensors can be disrupted by mist, steam, or large non-target particles (such as flying catkins). This can cause brief, dramatic fluctuations in the ambient visibility index (EVI) or the dust concentration (PM10), generating outliers that can interfere with the calculation and final decision-making of the air quality index (AQI). Therefore, filtering detected outliers can effectively remove brief, pulse-like interference (spikes) while better preserving the true trend of the data.

[0152] like Figure 9 Specifically, step S350 includes:

[0153] Step S351: Build an environmental visibility database, collect several spray painting workshop scene images with different lighting conditions, dust concentrations, and equipment layouts, and annotate each spray painting workshop scene image with a manually scored environmental visibility index (EVI);

[0154] Step S352: Anti-interference data enhancement is performed on the spray painting workshop scene image to form a simulated interference image. Specifically, Gaussian noise is added to the image; the atomization filter intensity is dynamically adjusted according to the dust concentration PM10 to generate a simulated dust interference image; the image brightness (±30%) and contrast (±20%) are randomly adjusted to simulate lighting changes.

[0155] Step S353: Build a deep learning model that simultaneously performs EVI regression prediction and key area segmentation, and design a multi-task loss function L = λ1·L_EVI+λ2·L_seg;

[0156] Where L_EVI is the mean square error loss of EVI regression, L_seg is the Dice loss of key region segmentation, which is calculated as 1-2×(predicted segmentation area ∩ true segmentation area) / (predicted area + true area); set the weight coefficients λ1=1.0, λ2=0.5;

[0157] Use simulated interference images to train the deep learning model until convergence;

[0158] Step S354: Input the environmental image information into the trained deep learning model, and synchronously output the EVI prediction value and the dust area segmentation map associated with PM10; based on the time sliding window statistics of EVI and PM10 instantaneous fluctuations, identify abnormal values ​​exceeding the preset threshold; filter the instantaneous anomaly detection results through the median filtering algorithm.

[0159] The above method constructs a single model to simultaneously perform two tasks: EVI regression prediction and key region segmentation. The key region segmentation task requires the model to identify areas in the image where dust concentration may be high (such as diffuse areas and accumulation areas near equipment). This design forces the model to focus not only on global features (for EVI) but also on local visual cues related to dust.

[0160] Introducing simulated instantaneous interference into the training data (such as simulated lens splash stains, simulated brief strong light, and simulated image blocks of non-dust particle interference) can improve the robustness of the model to real interference.

[0161] If the input image itself has instantaneous strong interference (such as strong light causing large-area overexposure) or non-dust occlusion, the EVI value predicted by the model may deviate significantly from the normal range (for example, if strong light causes the image to be too bright, the model may mistakenly judge it as extremely high visibility / EVI close to 1, but this is unrelated to dust concentration and is unreliable). More importantly, the dust segmentation map predicted by the model provides key clues: when real dust causes visibility to decrease, the segmentation map should show diffuse or localized dust areas; when instantaneous interference (such as lens stains, local strong light) causes EVI anomalies, the predicted dust segmentation map usually does not produce a high response in the interfered area, or it will produce isolated, oddly shaped responses that do not match the dust distribution pattern. This inconsistency between the EVI anomaly value and the expected pattern of the dust segmentation map is a key signal for identifying image-related anomalies.

[0162] Combining inconsistency analysis of model outputs (for EVI / image anomalies) with statistical anomaly detection of time series (for PM10 and EVI) allows for the identification of transient anomalies. Median filtering is then used to filter these detected anomalies, effectively removing brief pulse-like interference (spikes) while preserving the true trend of the data.

[0163] The above method corresponds to the intelligent environmental protection sensing device for the spraying operation environment, such as Figure 10 The intelligent environmental perception device also includes a data filtering module 6, which is disposed between the image processing module 2, the dust analysis module 3, and the data fusion module 4. The data filtering module 4 is configured to receive the environmental visibility index (EVI) from the image processing module 2 and the dust concentration (PM10) from the dust analysis module 3. The data filtering module 6 is also configured to detect and filter instantaneous abnormal values ​​in the environmental visibility index (EVI) and the dust concentration (PM10) using a deep learning algorithm.

[0164] The above-mentioned intelligent environmental protection perception method and device for the spraying operation environment can monitor the environmental visibility index EVI and air quality index AQI in real time and accurately, and can timely warn of safety risks brought about by reduced visibility (such as equipment operation hazards and personnel collision risks) and the harm to workers' health caused by excessive dust concentration, effectively reducing the accident rate, ensuring personnel safety and health, and helping enterprises to comply with regulations; accurate environmental perception data (EVI, PM10, AQI) provides a reliable basis for the intelligent regulation of the air supply and exhaust systems in the painting workshop, and can dynamically adjust the air volume according to actual environmental needs, avoid energy waste caused by excessive ventilation, and achieve energy saving and consumption reduction; stable environmental visibility (EVI) helps to ensure the clarity of operators' observation of the spraying effect, and indirectly assists in discovering spraying quality problems (such as sagging and orange peel).

[0165] The above-mentioned intelligent environmental protection perception method and device for the spray operation environment fully considered the challenges of lighting changes in the spraying workshop, complex equipment layout, special characteristics of particulate matter, etc. during design. Through the combination of AI algorithm and dual-wavelength sensing, it demonstrates strong environmental adaptability and practicality; it realizes integrated, real-time online monitoring of key indicators of the spray operation environment (visibility, dust concentration, comprehensive air quality), overcoming the shortcomings of traditional single-function equipment with scattered deployment and difficult data integration.

[0166] The present invention is not limited to the above-described embodiments. Any changes in shape or structure fall within the scope of protection of the present invention. The scope of protection of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention. Such changes and modifications shall fall within the scope of protection of the present invention.

Claims

1. A method for intelligent environmental protection perception of spraying operation environment, characterized in that: include: S100, collecting environmental image information of the spraying operation environment through a camera installed in the spraying operation environment; S200 , using a convolutional neural network in combination with a feature pyramid network to perform multi-scale feature fusion on the environmental image information to obtain an environmental visibility index (EVI) of the spraying operation environment.

2. The method for intelligent environmental protection perception of the spraying operation environment according to claim 1, characterized in that: The intelligent environmental protection perception method further includes: S300: Using a dust sensor to collect scattered light signals from the spraying operation environment, establishing a dust concentration inversion model based on the Mie scattering principle, and obtaining the dust concentration PM10.

3. The method for intelligent environmental protection perception of the spraying operation environment according to claim 2, characterized in that: The intelligent environmental protection perception method further includes: S400: Obtain the air quality index AQI according to the environmental visibility index EVI and the dust concentration PM10, AQI=W EVI ×(1-EVI)×100+(1-W EVI )×PM10 / 2,W EVI is the environmental visibility weight coefficient, 0.3≤W EVI ≤0.94; When PM10<100, W EVI Take the value as 0.8; when PM10<500, W EVI Take the value as 0.5; when PM10>500, W EVI The value is 0.

3.

4. The method for intelligent environmental protection perception of a spraying operation environment according to any one of claims 1 to 3, characterized in that: The step S200 includes: Step S201: pre-process the environmental image information to obtain an input image, where the input image is a 512×512 pixel RGB image; Step S202: The convolutional neural network uses ResNet-50 as the backbone network, inputs the input image into the convolutional neural network, and extracts three feature maps Stage 3, Stage 4, and Stage 5. The resolution of Stage 3 is 64×64×24, the resolution of Stage 4 is 32×32×40, and the resolution of Stage 5 is 16×16×80. Step S203: using the feature pyramid network and a top-down path, upsampling the Stage 5 by a factor of 2, and fusing the Stage 5, Stage 4, and Stage 3 step by step by element-by-element addition; Step S204: Use 1×1 convolution to adjust the channel lateral connection to generate multi-scale feature maps P3, P4 and P5, where the resolution of P3 is 64×64, the resolution of P4 is 32×32, and the resolution of P5 is 16×16; Step S205: Process the feature maps P3, P4, and P5 using formula (1) respectively. Ffusedi=Conv1×1(Pi)+α·UpSample(Ffusedi+1) (1) Among them, Ffused is the fused feature map, Conv1X1 is used to adjust the number of channels, α is the fusion weight coefficient, and i is 3, 4, or 5; Step S206: perform global average pooling on the fused feature map F_fused, output a 512-dimensional feature vector, and map it to the environmental visibility index EVI using formula (2) through the fully connected layer; EVI=σ(w T ·GAP(Ffused)+b) (2) Among them, σ is the Sigmoid activation function (output normalized to 0-1), w is the weight vector, GAP is the global average pooling of the image, Ffused is the multi-scale fused feature image, b is the bias term, and w and b are trainable parameters.

5. The method for intelligent environmental protection perception of the spraying operation environment according to claim 4, characterized in that: The step S200 further includes: Step S207: Establish dust concentration 0-2000 μg / m 3 The artificial visibility level is divided into artificial visibility level 1, artificial visibility level 2, and artificial visibility level 3. Artificial visibility level 1 is "good visibility, clear equipment outline", artificial visibility level 2 is "medium visibility, slightly blurred", and artificial visibility level 3 is "poor visibility, key equipment cannot be identified"; Step S208: Training the convolutional neural network and the feature pyramid network through the mean square error loss function so that the EVI matches the artificial visibility level. The EVI ≥ 0.8 corresponds to the first level of the artificial visibility level, the 0.5 ≤ EVI < 0.8 corresponds to the second level of the artificial visibility level, and the EVI < 0.5 corresponds to the third level of the artificial visibility level.

6. The method for intelligent environmental protection perception of a spraying operation environment according to claim 2 or 3, characterized in that: Step S300 includes: S301, the dust sensor is a dual-wavelength laser scattering sensor array, which emits a dual-wavelength laser beam into the spraying operation environment to form a scattered light signal of dust particles, and the main wavelength of the dual-wavelength laser scattering sensor array is 650nm and the auxiliary wavelength is 850nm; S302, receiving light signals scattered by dust particles through a signal conditioning circuit, and separating the scattered light of the 650nm channel and the 850nm channel respectively; S303. Convert the optical signal into an electrical signal through the 650nm channel and the 850nm channel, wherein the 650nm channel uses a transimpedance amplifier with a gain of 1 MΩ, and the 850nm channel uses a transimpedance amplifier with a gain of 500 kΩ. The bandwidth of the 650nm channel is 10 kHz, and the bandwidth of the 850nm channel is 10 kHz. S304, performing ADC sampling on the electrical signal with a sampling resolution of 16 bits and a sampling rate of 10 kSPS to generate a dual-wavelength digital signal; S305. Establish a single particle scattering intensity model based on Mie scattering theory, as shown in formula (3): Where I is the scattered light intensity, S is the Mie scattering complex amplitude function, which is numerically solved by the Bessel function; λ is the specific wavelength; θ is the scattering angle, d is the particle size, and m is the complex refractive index; S306, using a pre-built dual-wavelength lookup table LUT, performing particle size inversion on the digital signal by non-negative least squares method, wherein the LUT is generated by pre-calculating the scattering intensity of different d and m combinations at wavelengths of 650nm and 850nm, Among them, the particle size range d is 0.3-10 μm, the step size λ is 0.1 μm, and the refractive index m∈[1.3+0i,1.7+0.1i]; S307. Output the dust concentration PM10 according to the inversion result.

7. The method for intelligent environmental protection perception of the spraying operation environment according to claim 3, characterized in that: The intelligent environmental protection perception method also includes step S350, which is before step S400 and after step S300. Step S350 uses a deep learning algorithm to detect and filter instantaneous abnormal values ​​in the environmental visibility index EVI and dust concentration PM10.

8. The method for intelligent environmental protection perception of the spraying operation environment according to claim 7, characterized in that: The step S350 includes: Step S351: Build an environmental visibility database, collect a number of spray painting workshop scene images with different lighting conditions, dust concentrations, and equipment layouts, and annotate each of the spray painting workshop scene images with a manually scored environmental visibility index (EVI); Step S352: performing anti-interference data enhancement on the spray painting workshop scene image to form a simulated interference image; Step S353: Build a deep learning model that simultaneously performs EVI regression prediction and key area segmentation, and design a multi-task loss function L = λ1·L_EVI+λ2·L_seg; Where L_EVI is the mean square error loss of EVI regression, L_seg is the Dice loss of key region segmentation, which is calculated as 1-2×(predicted segmentation area ∩ true segmentation area) / (predicted area + true area); set the weight coefficients λ1=1.0, λ2=0.5; Using the simulated interference image to train the deep learning model until convergence; Step S354: Input the environmental image information into the trained deep learning model, and synchronously output the EVI prediction value and the dust area segmentation map associated with PM10; based on the time sliding window statistics of the instantaneous fluctuations of EVI and PM10, identify abnormal values ​​exceeding the preset threshold; filter the instantaneous abnormality detection results through the median filtering algorithm.

9. An intelligent environmental protection sensing device for spraying operation environment, characterized in that: The intelligent environmental protection perception device for the spraying operation environment uses the intelligent environmental protection perception method according to any one of claims 1 to 8, and the intelligent environmental protection perception device includes: An image acquisition module, the image acquisition module is used to acquire environmental image information of the spraying operation environment; An image processing module is used to perform multi-scale feature fusion on the environmental image information using a convolutional neural network combined with a feature pyramid network to obtain an environmental visibility index (EVI) of the spraying operation environment; A dust analysis module is used to collect scattered light signals from the spraying operation environment, establish a dust concentration inversion model based on the Mie scattering principle, and obtain the dust concentration PM10; A data fusion module is configured to receive the environmental visibility index EVI from the image processing module and the dust concentration PM10 from the dust analysis module, and fuse the environmental visibility index EVI and the dust concentration PM10 to obtain an air quality index AQI.

10. The intelligent environmental protection sensing device for spraying operation environment according to claim 9, characterized in that: The intelligent environmental protection perception device also includes a data filtering module, which is arranged between the image processing module, the dust analysis module and the data fusion module. The data filtering module is used to receive the environmental visibility index EVI of the image processing module and the dust concentration PM10 of the dust analysis module. The data filtering module is also used to use a deep learning algorithm to detect and filter instantaneous abnormal values ​​in the environmental visibility index EVI and dust concentration PM10.

Images