A high-pressure micro-mist dust removal system combined with a target detection algorithm

Abstract

The present invention belongs to the technical field of fault diagnosis, and discloses a high-pressure micro-mist dust removal system combined with a target detection algorithm, which includes a detection platform, a water cable for transporting water source, a high-pressure micro-mist integration system and high-pressure atomizing nozzles. The water cable for transporting water source is used for supplying water, and the water flows through the high-pressure micro-mist integration system and is transported to the corresponding high-pressure atomizing nozzles. The detection platform collects on-site images through an image acquisition module, and detects the position and range of dust through an on-site dust positioning detection algorithm, and then controls the high-pressure atomizing nozzles at the corresponding positions to spray, and the spraying range is controlled according to the detected dust range. The present invention effectively improves the detection accuracy of the position and range of dust, so as to be able to locate the specific position of the dust and guide the working position and range of spraying; and can accurately select the number, position and spraying range of the corresponding nozzles, realize precise dust reduction, and save equipment resources.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and relates to a high-pressure micro-mist dust removal system combined with a target detection algorithm. Background Art

[0002] In storage and transfer yards for powdered materials or yards with more dust, a large amount of dust is likely to occur, which not only affects the operation of the yard but also causes damage to the physical health of on-site workers. Therefore, dust removal equipment needs to be used to reduce dust and purify the environment. However, many existing yards need to frequently transport and discharge materials during working hours. Therefore, different degrees of dust will also be generated during the material transmission process, and it is difficult for manual control equipment to spray dust in a timely manner. Existing automated dust removal equipment has a high cost and a large equipment investment. Also, since dust itself is a relatively difficult-to-accurately identify small target, existing target detection methods are difficult to achieve the positioning detection of dust, and thus cannot accurately reduce dust. In order to achieve the dust reduction effect, only the spraying range can be expanded, resulting in inaccurate spraying range, waste of resources, and it is difficult to reduce the use cost. Summary of the Invention

[0003] The purpose of the present invention is to provide a high-pressure micro-mist dust removal system combined with a target detection algorithm, which is used to solve the technical problems in the prior art that automated dust removal equipment is difficult to achieve the positioning detection of dust, cannot accurately reduce dust, wastes resources, and is difficult to reduce the use cost.

[0004] The high-pressure micro-mist dust removal system combined with the target detection algorithm includes a detection platform, a water cable for transporting water source, a high-pressure micro-mist integration system, and high-pressure atomizing nozzles. The water cable for transporting water source is used for water supply, and the water flows through the high-pressure micro-mist integration system and is transported to the corresponding high-pressure atomizing nozzles. The detection platform collects on-site images through an image acquisition module, and detects the position and range of dust through an on-site dust positioning detection algorithm, and then controls the high-pressure atomizing nozzles at the corresponding positions to spray, and the spraying range is controlled according to the detected dust range.

[0005] The on-site dust positioning detection algorithm uses a lightweight network YOLOx model as the target detection model. The network YOLOx model includes a backbone part, a bottleneck network, and a target detection head. The backbone part is a CSPDarknet structure, and the operation is convolution to different degrees to extract low-dimensional and high-dimensional feature information of the input image; this model uses a convolutional layer with a residual structure; the bottleneck network uses an SPP structure and performs max pooling using pooling kernels of different sizes; after the feature information is sent to the YoLoHead module, the classification prediction and anchor box regression tasks are realized through convolution.

[0006] Preferably, in the feature extraction stage, the SPP module converts the residual edge into a global context awareness model.

[0007] Preferably, the target detection model uses a feature fusion module which connects feature information of different sizes and dimensions in an anti-convolution and down-sampling manner, and finally sends the feature information obtained from the foregoing processing to the YoLoHead module.

[0008] Preferably, the usage method of the high-pressure micro-mist dust removal system includes the following steps:

[0009] Step 1: Collect images of different degrees of dust conditions during the feeding process at the stockpiling site and make them into corresponding image datasets;

[0010] Step 2: Perform a tagging operation on the image dataset to label the position range and type information of the dust to be measured corresponding to the images;

[0011] Step 3: Randomly divide the labeled image dataset into a training set and a test set, and use the training set to train the target detection model;

[0012] Step 4: Observe whether the loss function of the target detection model converges during the training process. If it converges to the best, proceed to the next step; otherwise, continue to train with other labeled images in the training set;

[0013] Step 5: Save the best weight parameters obtained from the training to obtain a trained target detection model;

[0014] Step 6: Use the trained on-site dust positioning detection algorithm to detect dust. While turning on the real-time screen of the camera, load the target detection model to detect the on-site screen transmitted by the camera in real time;

[0015] Step 7: If dust is generated at the feeder opening, the detection platform measures the position and range of the dust, and based on the measured dust position, turns on the corresponding high-pressure atomizing nozzle to accurately control the spraying range to achieve precise dust removal.

[0016] Preferably, in Step 6, the camera installed at the stockpiling site is used as the graphic acquisition module, and the detection platform is equipped with an open system of the camera, and the on-site dust positioning detection algorithm is deployed into the development framework of the camera.

[0017] Preferably, in Step 7, after detecting dust, the detection platform will turn on the high-pressure micro-mist integration system, turn on the corresponding high-pressure atomizing nozzle based on the measured dust position, and accurately control the spraying range based on the measured dust range, so as to achieve relatively precise dust removal.

[0018] The present invention has the following advantages: The high-pressure micro-mist dust removal system provided by the present invention, on the one hand, uses an improved target detection model to detect the dust situation at various positions during the material transportation process in real time; effectively improves the detection accuracy of the dust position range, so as to locate the specific position of the dust and guide the working position and range of the spray; on the other hand, utilizes the high-pressure micro-mist integration system and multiple high-pressure atomizing nozzles arranged on-site to accurately select the number, position and spraying range of the corresponding nozzles, realize precise dust reduction, save equipment resources, and reduce the cost of dust removal. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic structural diagram of a high-pressure micro-mist dust removal system combining a target detection algorithm in the present invention.

[0020] Figure 2 It is a network structure diagram of the target detection model adopted by the present invention.

[0021] Figure 3 It is a schematic diagram of the deconvolution operation process in the present invention.

[0022] Figure 4 It is a schematic structural diagram of the improved SPP module in the present invention.

[0023] Figure 5 It is a flowchart of the usage method of the high-pressure micro-mist dust removal system in the present invention.

[0024] The reference numerals in the drawings include: 1, detection platform; 2, camera; 3, high-pressure atomizing nozzle; 4, high-pressure micro-mist integration system; 5, water cable for water supply. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following will further describe in detail the specific embodiments of the present invention with reference to the drawings through the description of the embodiments, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0026] As Figures 1-5 shown, the present invention discloses a high-pressure micro-mist dust removal system combining a target detection algorithm, including a detection platform 1, a water cable for water supply 5, a high-pressure micro-mist integration system 4 and high-pressure atomizing nozzles 3. The water cable for water supply 5 is used for water supply, and the water flows through the high-pressure micro-mist integration system 4 and is transported to the corresponding high-pressure atomizing nozzles 3. The detection platform 1 collects on-site images through an image acquisition module, and detects the position and range of dust through an on-site dust positioning detection algorithm, and then controls the high-pressure atomizing nozzles 3 at the corresponding positions to spray, and the spraying range is controlled according to the detected dust range.

[0027] The on-site dust positioning detection algorithm uses the lightweight network YOLOx model as the object detection model. The network YOLOx model includes a backbone part, a Bottleneck network, and a YoloHead. The backbone part is a CSPDarknet structure, and its main operation is convolution to varying degrees, extracting low-dimensional and high-dimensional feature information of the input image. To prevent information loss during the feature extraction process, the model uses a convolutional layer with a residual structure. At the same time, to obtain feature maps with different receptive fields, the Bottleneck network uses an SPP structure and performs max pooling using pooling kernels of different sizes. To obtain information features of different dimensions and scales, the model uses a feature fusion module, connecting feature information of different sizes and dimensions through transposed convolution and downsampling. Finally, the feature information obtained from the above processing is sent into the YoLoHead module, and the category prediction and anchor box regression tasks are achieved through convolution.

[0028] In the feature extraction stage, the object detection model introduces an SPP module (SPP Bottleneck), which uses pooling operations with different-sized convolutional kernels to obtain feature maps with different receptive fields. In the original SPP network, the SPP module connects residual edges, which can prevent information loss but also generates redundant feature information. To prevent this, we improve this module by converting the residual edge into a global context awareness model, preventing feature information loss without generating redundant information.

[0029] In this solution, when connecting feature maps of different levels, the traditional upsampling method is changed to a transposed convolution operation (UpSampling2D), so as to better restore the sizes of feature maps of different sizes and enable better connection of feature information between different levels. And this method draws on the idea of the residual network and also parallelly retains its corresponding feature structure during the process of fusing feature maps of different sizes.

[0030] The detection platform 1 uses the on-site dust positioning detection algorithm obtained through the above improvements, and uses the camera 2 installed at the stockpiling site as the graphic acquisition module. The detection platform 1 is paired with the open system of the camera 2, deploys the on-site dust positioning detection algorithm into the development framework of the camera 2, loads the object detection model while opening the real-time screen of the camera 2, and detects the on-site screen transmitted in real time by the camera 2.

[0031] The camera 2 collects dust images of different degrees at the stockpiling site as the dataset for model training. Then, the tagging task is completed manually to determine the position information and category information of the dust in the images. The dataset is randomly divided into a training set and a test set. The training set data is fed into the object detection model for training, enabling the model to learn the features and position information of the dust in the images. When the model training reaches the best convergence state, the parameter weights of the model at this time are saved, and these parameter weights are used to guide the model to detect the target object.

[0032] During actual application, if dust is generated at the material discharge machine opening, the detection platform 1 detects the position and range of the dust based on the trained on-site dust positioning detection algorithm. At this time, the dust position will be displayed on the monitor. After detecting the dust, the detection platform 1 will activate the high-pressure micro-mist integration system 4, open the corresponding high-pressure atomizing nozzles 3 based on the measured dust position, and accurately control the spraying range based on the measured dust range, thereby achieving relatively precise dust removal.

[0033] Such as Figure 5 shown, the usage method of the high-pressure micro-mist dust removal system can be summarized into the following steps:

[0034] Step 1: Collect images of different degrees of dust conditions during the material discharging process at the stockpiling site and make them into a corresponding image dataset.

[0035] Step 2: Perform a tagging operation on the image dataset to label the position range and type information of the dust to be measured corresponding to the images.

[0036] Step 3: Randomly divide the labeled image dataset into a training set and a test set, and use the training set to train the object detection model.

[0037] Step 4: Observe whether the loss function of the object detection model converges during the training process. If it converges to the best, proceed to the next step; otherwise, continue to train with other labeled images in the training set.

[0038] Step 5: Save the best weight parameters obtained from training to obtain a trained object detection model.

[0039] Step 6: Use the trained on-site dust positioning detection algorithm to detect dust. While opening the real-time screen of the camera 2, load the object detection model to detect the on-site screen transmitted by the camera 2 in real time.

[0040] Step 7: If dust is generated at the material discharge machine opening, the detection platform 1 measures the position and range of the dust, opens the corresponding high-pressure atomizing nozzles 3 based on the measured dust position, and accurately controls the spraying range to achieve precise dust removal.

[0041] Through the high-pressure micro-mist dust removal system provided by the present invention, on the one hand, an improved target detection model is used to detect the dust situation at various positions during the material transmission process in real time; the detection accuracy of the dust position range is effectively improved, so that the specific position of the dust can be located to guide the working position and range of spraying; on the other hand, the high-pressure micro-mist integrated system 4 and multiple high-pressure atomizing nozzles 3 arranged on site are used to accurately select the number, position and spraying range of the corresponding nozzles, realizing precise dust reduction, saving equipment resources and reducing the cost of dust removal.

[0042] The present invention has been described exemplarily above in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above-mentioned manner. As long as various non-substantive improvements are made by adopting the inventive concept and technical solution of the present invention, or the inventive concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A high-pressure micro-mist dust removal system combined with a target detection algorithm, characterized in that: It includes a detection platform (1), a water cable for water supply (5), a high-pressure micro-mist integration system (4) and high-pressure atomizing nozzles (3). The water cable for water supply (5) is used for water supply. The water flows through the high-pressure micro-mist integration system (4) and is delivered to the corresponding high-pressure atomizing nozzles (3). The detection platform (1) collects on-site images through an image acquisition module, and detects the position and range of dust through an on-site dust positioning detection algorithm, and then controls the high-pressure atomizing nozzles (3) corresponding to the position to spray. The spraying range is controlled according to the detected dust range. The on-site dust positioning detection algorithm uses the lightweight network YOLOx model as the object detection model. The network YOLOx model includes a backbone part, a bottleneck network and an object detection head. The backbone part is a CSPDarknet structure, and the operation is convolution to different degrees to extract low-dimensional and high-dimensional feature information of the input image. This model uses a convolutional layer with a residual structure. The bottleneck network uses an SPP module and performs max pooling with pooling kernels of different sizes. After the feature information is sent into the YoLoHead module, the category prediction and anchor box regression tasks are realized through convolution. The usage method of the high-pressure micro-mist dust removal system includes the following steps: Step 1: Collect images of different degrees of dust conditions during the feeding process at the stockpiling site and make them into corresponding image datasets. Step 2: Perform a tagging operation on the image dataset to label the position range and type information of the dust to be measured corresponding to the image. Step 3: Randomly divide the labeled image dataset into a training set and a test set, and use the training set to train the object detection model. Step 4: Observe whether the loss function of the object detection model converges during the training process. If it converges to the best, proceed to the next step; otherwise, continue to train with other labeled images in the training set. Step 5: Save the best weight parameters obtained from the training to obtain a trained object detection model. Step 6: Use the trained on-site dust positioning detection algorithm to detect dust. While opening the real-time screen of the camera (2), load the object detection model and detect the on-site screen transmitted by the camera (2) in real time. Step 7: If dust is generated at the feeder opening, the detection platform (1) measures the position and range of the dust, opens the corresponding high-pressure atomizing nozzles (3) based on the measured dust position, and accurately controls the spraying range to achieve precise dust removal. In Step 6, the camera (2) installed at the stockpiling site is used as the graphic acquisition module. The detection platform (1) is paired with the open system of the camera (2), and the on-site dust positioning detection algorithm is deployed into the development framework of the camera (2). In Step 7, after the detection platform (1) detects dust, it will turn on the high-pressure micro-mist integration system (4), open the corresponding high-pressure atomizing nozzles (3) based on the measured dust position, and accurately control the spraying range based on the measured dust range, so as to achieve precise dust removal.

2. The high-pressure micro-mist dust removal system combined with the target detection algorithm according to claim 1, wherein: In the feature extraction stage, the SPP module converts the residual edge into a global context awareness model.

3. The high-pressure micro-mist dust removal system incorporating a target detection algorithm according to claim 1 or 2, characterized in that: The target detection model uses a feature fusion module, which connects feature information of different sizes and dimensions in an anti-convolution and down-sampling manner, and finally sends the aforementioned feature information obtained from the processing into the YoLoHead module.

Images