Intelligent bin unblocking method and system based on visual identification and air cannon linkage

Abstract

The embodiment of the invention provides a warehouse body intelligent unblocking method and system based on visual identification and air cannon linkage, and the method comprises the steps: collecting images in a warehouse in real time through a plurality of high-definition anti-explosion cameras disposed in the warehouse, carrying out the fusion processing of the collected images in the warehouse, and constructing a warehouse three-dimensional dynamic model; based on the warehouse three-dimensional dynamic model, key features are extracted by adopting a neural network; taking the extracted key features as input, analyzing associated parameters through a time sequence of a long-short-term memory network, establishing a congestion cause knowledge base, and labeling congestion types in a classified manner; constructing a congestion risk grading model based on the congestion cause library and a machine learning algorithm; and activating a corresponding air cannon based on the level corresponding to the three-level early warning and a visual positioning matching result, dynamically adjusting injection parameters, and executing unblocking operation according to an optimized time sequence. According to the embodiment of the invention, closed-loop intelligence from visual perception, reason modeling, artificial intelligence decision making to accurate execution is constructed.

Description

Technical Field

[0001] This application belongs to the technical field of warehouse unblocking, specifically involving a warehouse intelligent unblocking method and system based on visual recognition and air cannon linkage. Background Technology

[0002] In industries such as power, mining, building materials, grain, and chemicals, storage silos (including coal silos, cement silos, ore silos, grain silos, and powder silos) are core facilities for material storage and transportation. Storage silo blockages (such as adhesion to walls, clumping and bridging, and particle retention) are a common problem that has long plagued the industry.

[0003] Traditional methods of clearing blockages mainly rely on manual intervention and tools such as jacks and hammers. This is labor-intensive, cannot reach high-level blockage points, and poses high safety risks.

[0004] Common methods for clearing blockages include air cannons, vibratory hammers, and scraper machines. However, these devices cannot achieve precise clearing from multiple angles and all directions; sometimes, the more you knock, the more compacted the blockage becomes, and the more you blow, the more severe the blockage becomes. While the shape of warehouses and the goods loaded inside may differ across industries and warehouses, the blockage patterns are generally similar, including: wall-adhesive blockages, bridging / bridging blockages, agglomerated blockages, and rat-hole / funnel-flow blockages.

[0005] Currently, there is no solution that can study and resolve the aforementioned common problems. Summary of the Invention

[0006] This application proposes a method and system for intelligent unblocking of cargo compartments based on visual recognition and air cannon linkage, in order to overcome the shortcomings of the aforementioned prior art.

[0007] According to a first aspect of the embodiments of this application, a method for intelligent unblocking of a cargo compartment based on visual recognition and air cannon linkage is provided, including: Multiple high-definition explosion-proof cameras deployed inside the storage warehouse are used to collect images of the warehouse in real time. The collected images are then fused together to construct a three-dimensional dynamic model of the storage warehouse. Based on the aforementioned three-dimensional dynamic model of the storage warehouse, key features are extracted using a neural network; The extracted key features are used as input, and the correlation parameters are analyzed through time series analysis of long short-term memory network. A knowledge base of blockage causes is established and the blockage types are classified and labeled. Based on the aforementioned congestion cause library and machine learning algorithm, a congestion risk classification model is constructed, and a three-level early warning is generated; Based on the matching result of the level and visual positioning corresponding to the three-level early warning, the corresponding air cannon is activated, the spray parameters are dynamically adjusted, and the blockage clearing operation is performed according to the optimized timing sequence.

[0008] In some embodiments, the multiple high-definition explosion-proof cameras include fisheye lenses and planar lenses, and the process includes: [Further details to be added] The fisheye lens is used to cover the entire top of the silo, and the planar lens is used to focus on the cone section and the discharge port, thus enhancing the comprehensiveness of image acquisition in a combined manner.

[0009] In some implementations, the process of fusing the acquired images inside the warehouse and constructing a three-dimensional dynamic model of the warehouse includes: The collected images inside the warehouse are processed through multi-view image fusion to generate a three-dimensional dynamic model of the warehouse, and the accuracy of the three-dimensional dynamic model of the warehouse is controlled to meet the preset error range.

[0010] In some implementations, the key features include material packing morphology, particle agglomeration degree, and bin wall adhesion degree, and after extracting the key features using a neural network, the following are included: When the material accumulation angle in the material accumulation pattern exceeds a preset threshold, it is marked and warned as a risk of wall adhesion; otherwise, the current state is maintained. When the agglomeration area of ​​the particles exceeds a set threshold, it is marked and an early warning is issued as a sign of stubborn blockage; otherwise, the current state is maintained. The percentage of pixels in the silo wall adhesion layer is calculated based on the edge detection algorithm to dynamically identify early signs of blockage.

[0011] In some implementations, the correlation parameters include historical data, material parameters, and environmental parameters, and the time series analysis of the correlation parameters through long short-term memory networks includes: By analyzing historical data such as the frequency of historical blockages, material moisture content, particle size distribution, and warehouse humidity through time series analysis of long short-term memory networks, a parameter set is used to construct a multi-dimensional causal knowledge base.

[0012] In some implementations, the three-level early warning system includes an initial adhesion stage, a developmental bridging stage, and a mature, completely blocked stage. The generation of the three-level early warning system includes: Based on the clogging probability of the clogging risk classification model, the initial adhesion stage, the development bridging stage, and the mature complete clogging stage are generated. A pretreatment procedure is triggered during the initial bonding stage to enable early intervention.

[0013] In some implementations, the visual positioning matching result is the coordinate result of the blockage area determined by three-dimensional reconstruction through the three-dimensional dynamic model of the storage warehouse, which is used to match the blockage clearing location.

[0014] In some embodiments, the dynamic adjustment of injection parameters includes: The required pressure is dynamically calculated based on the adhesion layer thickness in the bin wall adhesion degree, and the nozzle shape is changed by changing the spiral variable diameter pipe head to adjust the airflow diffusion angle so that the energy matches the peeling strength.

[0015] In some implementations, the optimized timing for performing the deblocking operation includes: The timing optimization is performed according to the interval blowing order of edge before center; The blockage is cleared by utilizing the superposition effect of airflow to expand the stripping area.

[0016] According to a first aspect of the embodiments of this application, a smart unblocking system for cargo compartments based on visual recognition and air cannon linkage is provided, comprising: The 3D model building module is used to collect images inside the storage warehouse in real time by deploying multiple high-definition explosion-proof cameras inside the warehouse, and to fuse the collected images inside the warehouse to build a 3D dynamic model of the storage warehouse. The key feature extraction module is used to extract key features based on the three-dimensional dynamic model of the storage warehouse using a neural network. The blockage cause statistics module is used to take the extracted key features as input, analyze the correlation parameters through time series analysis of long short-term memory network, and establish a blockage cause knowledge base and classify and label blockage types. The risk warning module is used to construct a congestion risk classification model based on the congestion cause library and machine learning algorithm, and generate a three-level warning. The blockage clearing control and execution module is used to activate the corresponding air cannon based on the matching result of the level corresponding to the three-level warning and the visual positioning, dynamically adjust the injection parameters, and execute the blockage clearing operation according to the optimized timing sequence.

[0017] The beneficial effects of the intelligent unblocking method and system for cargo compartments based on visual recognition and air cannon linkage in this application embodiment include at least the following: This application's embodiments accurately reconstruct the spatial distribution of materials using a 3D model, providing a geometric benchmark for blockage location. This enables real-time, blind-spot-free monitoring of blockage areas within the storage facility, eliminating blind spots in manual inspections, improving the reliability of blockage point coordinate calculations, and avoiding deep misjudgments caused by traditional 2D identification. Early identification of potential blockage risks (such as tiny adhesion points) facilitates proactive warnings. Automatic quantification of blockage characteristics reduces reliance on manual experience, standardizes feature extraction criteria, and adapts to different material characteristics (coal powder / cement / grain, etc.). Automatic labeling of blockage types provides targeted treatment guidelines. By linking historical data and environmental variables with multiple source parameters, the root causes of blockages are revealed, forming an iteratively optimizable knowledge base and enhancing the system's adaptability. Quantifying the blockage development stage using a probabilistic model avoids excessive or delayed intervention. A three-level early warning mechanism matches differentiated response strategies, reducing the frequency of ineffective blockage clearing. Activating the nearest air cannon based on coordinate matching shortens the airflow path and improves energy utilization. Dynamic parameter tuning ensures energy matching with blockage intensity, preventing damage to the storage facility. Optimizing the injection sequence and airflow superposition effect through timing optimization expands the coverage area of ​​a single blockage clearing operation. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the intelligent unblocking method for a cargo compartment based on visual recognition and air cannon linkage, as described in an embodiment of this application. Figure 2 This is a structural diagram showing the arrangement of the camera and air cannon in the cargo compartment according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the intelligent unblocking system for a cargo compartment based on visual recognition and air cannon linkage, according to an embodiment of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technology of this application, the following detailed description of this application is provided in conjunction with the accompanying drawings and specific embodiments.

[0020] The embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The detailed description of the following embodiments and the accompanying drawings are used to illustrate the principles of this application by way of example, but should not be used to limit the scope of this application, that is, this application is not limited to the described embodiments.

[0021] This application discloses an intelligent unblocking method for warehouses based on visual recognition and air cannon linkage. This method is executed using an intelligent unblocking system for warehouses based on visual recognition and air cannon linkage. The method includes steps 110-150 and is particularly suitable for warehouse blockage control in industries such as thermal power generation, mining, and building materials.

[0022] Step 110: Real-time images of the storage warehouse are collected by multiple high-definition explosion-proof cameras deployed inside the warehouse, and the collected images are fused to construct a three-dimensional dynamic model of the storage warehouse.

[0023] In some embodiments, the warehouse camera (hereinafter referred to as "camera") in this application embodiment, that is, the multiple high-definition explosion-proof cameras, may include a fisheye lens and a planar lens.

[0024] In some implementations, before acquiring real-time images of the warehouse using multiple high-definition explosion-proof cameras deployed within the warehouse, the method includes: using the fisheye lens to cover the panoramic view of the warehouse roof and using the planar lens to focus on the cone section and the unloading port, thereby enhancing the comprehensiveness of image acquisition in a combined manner.

[0025] In some implementations, the step of fusing the acquired images inside the warehouse and constructing a three-dimensional dynamic model of the warehouse includes: generating the three-dimensional dynamic model of the warehouse by multi-view image fusion processing of the acquired images inside the warehouse, and controlling the accuracy of the three-dimensional dynamic model of the warehouse to meet a preset error range.

[0026] For example, see Appendix Figure 2 As shown, in this embodiment of the application, 3-6 high-definition explosion-proof cameras 210 are deployed (adjusted according to the diameter of the storage silo from 3-20m). A combination of fisheye lens and planar lens is used (the fisheye lens covers the panoramic view of the silo top, and the planar lens focuses on the cone section and the unloading port). Images inside the silo are collected in real time at 25 frames / second, and a three-dimensional dynamic model of the storage silo is constructed through image fusion technology (accuracy up to ±1.5cm).

[0027] Step 120: Based on the three-dimensional dynamic model of the storage warehouse, key features are extracted using a neural network.

[0028] In some implementations, the key features include material accumulation morphology, particle agglomeration degree, and silo wall adhesion degree. After extracting the key features using a neural network, the process includes: when the material accumulation angle in the material accumulation morphology exceeds a preset threshold, marking it as a risk of wall adhesion; otherwise, maintaining the current state; when the agglomeration area in the particle agglomeration degree exceeds a set threshold, marking it as a precursor to stubborn blockage; otherwise, maintaining the current state; and calculating the percentage of pixels in the silo wall adhesion layer based on an edge detection algorithm to dynamically identify early signs of blockage.

[0029] For example, embodiments of this application can utilize an improved ResNet-50 network to extract key features such as material stacking morphology (e.g., stacking angle > 45° to warn of wall adhesion risk), particle agglomeration degree (e.g., agglomeration area > 0.3m² marked as a precursor to stubborn blockage), and silo wall adhesion thickness (calculated based on edge detection algorithm to determine the percentage of adhesive layer pixels).

[0030] Step 130: Use the extracted key features as input, analyze the correlation parameters through time series analysis of a long short-term memory network (LSTM), and establish a knowledge base for the causes of blockages and classify and label the types of blockages.

[0031] In some embodiments, the correlation parameters include historical data, material parameters, and environmental parameters. It is characterized in that the correlation of parameters through time series analysis of the long short-term memory network includes: analyzing the historical blockage frequency in historical data, the moisture content of the material in the material parameters, the particle size distribution in the material parameters, and the humidity parameter in the warehouse in the environmental parameters through time series analysis of the long short-term memory network, so as to constitute a parameter set for generating a multi-dimensional cause knowledge base.

[0032] For example, in the embodiments of the present application, through time series analysis of the long short-term memory network, historical blockage data (such as the blockage frequency at the same location in the past 30 days), material parameters (moisture content, particle size distribution), and environmental parameters (humidity in the warehouse > 60% triggers a high adhesion warning) are correlated to establish a knowledge base for the causes of blockages, automatically label 8 types of blockage types such as "wet adhesion to the wall", "uneven particle size bridging", "moisture absorption and caking", etc., and count the proportion of each cause.

[0033] Step 140: Build a blockage risk grading model based on the blockage cause library and machine learning algorithms, and generate a three-level warning.

[0034] In some embodiments, the three-level warning includes an initial adhesion stage, a development stage bridging stage, and a mature stage complete blockage stage.

[0035] In some embodiments, the generation of the three-level warning includes: generating the initial adhesion stage, the development stage bridging stage, and the mature stage complete blockage stage based on the blockage probability of the blockage risk grading model; triggering a pre-disposal program in the initial adhesion stage to achieve early intervention.

[0036] For example, in the embodiments of the present application, through early研判 and dynamic decision-making of artificial intelligence (AI), a blockage risk grading model is generated. For example, a three-layer warning system is constructed based on the XGBoost algorithm, and the blockage probability P = parameter f (visual features, material parameters, and historical data) is calculated in real time. Among them, when 0.3 < P < 0.6, it is determined as the initial adhesion stage (such as dot adhesion on the warehouse wall), 0.6 < P < 0.9 is the development stage bridging stage (forming a local material bridge), and P > 0.9 is the mature stage complete blockage. The system will start a pre-disposal program when P < 0.3 to achieve "preventive treatment" type blockage clearing.

[0037] Step 150: Activate the corresponding air cannon based on the level corresponding to the three-level warning and the visual positioning matching result, dynamically adjust the spraying parameters, and perform the blockage clearing operation according to the optimized time sequence.

[0038] In some implementations, the visual positioning matching result is the coordinate result of the blockage area determined by three-dimensional reconstruction through the three-dimensional dynamic model of the storage warehouse, which is used to match the blockage clearing location. For example, the centroid coordinates (X, Y, Z) of the adhesion area are determined by visual three-dimensional reconstruction, and air cannons within ±0.5m of these coordinates are preferentially matched. For example, if the coordinates are (3, 4, 5), the third air cannon on the fourth layer is activated.

[0039] In some embodiments, the dynamic adjustment of injection parameters includes: dynamically calculating the required pressure based on the adhesion layer thickness (in centimeters) in the bin wall adhesion degree to ensure that the energy is precisely matched to the adhesion strength; and adjusting the airflow diffusion angle by changing the nozzle shape through a spiral variable diameter nozzle head to match the peeling strength.

[0040] In some embodiments, the optimized timing of the unblocking operation includes: performing the optimized timing according to the interval of blowing from the edge before the center (0.2 seconds interval); and expanding the stripping area by utilizing the airflow superposition effect to perform the unblocking.

[0041] For example, refer again to the appendix Figure 2 As shown, the air cannon 220's rotating mechanism uses a 360° omnidirectional rotating base to precisely control the injection direction. The nozzle uses a spiral variable diameter tube head, adjusting the airflow direction and diffusion angle by changing the nozzle shape (convergence / expansion). When the system detects that the adhesion area growth rate is >5% / min (initial blockage characteristics), it automatically triggers dynamic adjustment of injection parameters and executes unblocking operations according to the optimized timing sequence.

[0042] This application's embodiments accurately reconstruct the spatial distribution of materials using a 3D model, providing a geometric benchmark for blockage location. This enables real-time, blind-spot-free monitoring of blockage areas within the storage facility, eliminating blind spots in manual inspections, improving the reliability of blockage point coordinate calculations, and avoiding deep misjudgments caused by traditional 2D identification. Early identification of potential blockage risks (such as tiny adhesion points) facilitates proactive warnings. Automatic quantification of blockage characteristics reduces reliance on manual experience, standardizes feature extraction criteria, and adapts to different material characteristics (coal powder / cement / grain, etc.). Automatic labeling of blockage types provides targeted treatment guidelines. By linking historical data and environmental variables with multiple source parameters, the root causes of blockages are revealed, forming an iteratively optimizable knowledge base and enhancing the system's adaptability. Quantifying the blockage development stage using a probabilistic model avoids excessive or delayed intervention. A three-level early warning mechanism matches differentiated response strategies, reducing the frequency of ineffective blockage clearing. Activating the nearest air cannon based on coordinate matching shortens the airflow path and improves energy utilization. Dynamic parameter tuning ensures energy matching with blockage intensity, preventing damage to the storage facility. Optimizing the injection sequence and airflow superposition effect through timing optimization expands the coverage area of ​​a single blockage clearing operation.

[0043] This application also discloses an intelligent unblocking system for cargo compartments based on visual recognition and air cannon linkage. (See attached document.) Figure 3 As shown, the system includes: a 3D model building module 310, a key feature extraction module 320, a blockage cause statistics module 330, a risk warning building module 340, and a blockage clearing control and execution module 350.

[0044] The 3D model building module 310 is used to collect images inside the storage warehouse in real time by using multiple high-definition explosion-proof cameras deployed inside the warehouse, and to fuse the collected images inside the warehouse to build a 3D dynamic model of the storage warehouse.

[0045] The key feature extraction module 320 is used to extract key features based on the three-dimensional dynamic model of the storage warehouse using a neural network.

[0046] The blockage cause statistics module 330 is used to take the extracted key features as input, analyze the correlation parameters through the time series analysis of the long short-term memory network, and establish a blockage cause knowledge base and classify and label the blockage types.

[0047] The risk warning construction module 340 is used to construct a congestion risk classification model based on the congestion cause library and machine learning algorithm, and generate a three-level warning.

[0048] The blockage clearing control and execution module 350 is used to activate the corresponding air cannon based on the matching result of the level and visual positioning corresponding to the three-level warning, dynamically adjust the injection parameters and execute the blockage clearing operation according to the optimized timing sequence.

[0049] This application also discloses an electronic device that integrates the above-described intelligent unblocking system for cargo compartments based on visual recognition and air cannon linkage, for implementing the above-described intelligent unblocking method for cargo compartments based on visual recognition and air cannon linkage.

[0050] This application embodiment constructs a closed-loop intelligent system from visual perception, cause modeling, artificial intelligence decision-making to precise execution. It deeply analyzes the causes of storage warehouse blockage through visual neural networks and realizes early prediction and precise intervention of blockage risk based on artificial intelligence algorithms.

[0051] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of this application, and this application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this application, and these modifications and improvements are also considered to be within the scope of protection of this application.

Claims

1. A method for intelligent unblocking of cargo compartments based on visual recognition and air cannon linkage, characterized in that, include: Multiple high-definition explosion-proof cameras deployed inside the storage warehouse are used to collect images of the warehouse in real time. The collected images are then fused together to construct a three-dimensional dynamic model of the storage warehouse. Based on the aforementioned three-dimensional dynamic model of the storage warehouse, key features are extracted using a neural network; The extracted key features are used as input, and the correlation parameters are analyzed through time series analysis of long short-term memory network. A knowledge base of blockage causes is established and the blockage types are classified and labeled. Based on the aforementioned congestion cause library and machine learning algorithm, a congestion risk classification model is constructed, and a three-level early warning is generated; Based on the matching result of the level and visual positioning corresponding to the three-level early warning, the corresponding air cannon is activated, the spray parameters are dynamically adjusted, and the blockage clearing operation is performed according to the optimized timing sequence.

2. The method according to claim 1, wherein the plurality of high-definition explosion-proof cameras include a fisheye lens and a planar lens, characterized in that, Before collecting real-time images of the warehouse using multiple high-definition explosion-proof cameras deployed within it, the following steps were taken: The fisheye lens is used to cover the entire top of the silo, and the planar lens is used to focus on the cone section and the discharge port, thus enhancing the comprehensiveness of image acquisition in a combined manner.

3. The method according to claim 1, characterized in that, The process of fusing the collected images inside the warehouse and constructing a three-dimensional dynamic model of the warehouse includes: The collected images inside the warehouse are processed through multi-view image fusion to generate a three-dimensional dynamic model of the warehouse, and the accuracy of the three-dimensional dynamic model of the warehouse is controlled to meet the preset error range.

4. The method according to claim 1, characterized in that, The key features include material packing morphology, particle agglomeration degree, and bin wall adhesion degree, which, after extracting the key features using a neural network, include: When the material accumulation angle in the material accumulation pattern exceeds a preset threshold, it is marked and warned as a risk of wall adhesion; otherwise, the current state is maintained. When the agglomeration area of ​​the particles exceeds a set threshold, it is marked and an early warning is issued as a sign of stubborn blockage; otherwise, the current state is maintained. The percentage of pixels in the silo wall adhesion layer is calculated based on the edge detection algorithm to dynamically identify early signs of blockage.

5. The method according to claim 1, wherein the associated parameters include historical data, material parameters, and environmental parameters, characterized in that, The time series analysis parameters obtained through the Long Short-Term Memory Network include: By analyzing historical data such as the frequency of historical blockages, material moisture content, particle size distribution, and warehouse humidity through time series analysis of long short-term memory networks, a parameter set is used to construct a multi-dimensional causal knowledge base.

6. The method according to claim 1, wherein the three-level early warning includes an initial adhesion stage, a development bridging stage, and a mature complete blockage stage, characterized in that, The generation of the three-level early warning includes: Based on the clogging probability of the clogging risk classification model, the initial adhesion stage, the development bridging stage, and the mature complete clogging stage are generated. A pretreatment procedure is triggered during the initial bonding stage to enable early intervention.

7. The method according to claim 1, characterized in that, The visual positioning matching result is the coordinate result of the blockage area determined by the three-dimensional reconstruction of the storage warehouse's three-dimensional dynamic model, which is used to match the blockage clearing location.

8. The method according to claim 4, characterized in that, The dynamic adjustment of injection parameters includes: The required pressure is dynamically calculated based on the adhesion layer thickness in the bin wall adhesion degree, and the nozzle shape is changed by changing the spiral variable diameter pipe head to adjust the airflow diffusion angle so that the energy matches the peeling strength.

9. The method according to claim 1, characterized in that, The optimized timing for performing the congestion clearing operation includes: The timing optimization is performed according to the interval blowing order of edge before center; The blockage is cleared by utilizing the superposition effect of airflow to expand the stripping area.

10. A smart unblocking system for cargo compartments based on visual recognition and air cannon linkage, characterized in that, include: The 3D model building module is used to collect images inside the storage warehouse in real time by deploying multiple high-definition explosion-proof cameras inside the warehouse, and to fuse the collected images inside the warehouse to build a 3D dynamic model of the storage warehouse. The key feature extraction module is used to extract key features based on the three-dimensional dynamic model of the storage warehouse using a neural network. The blockage cause statistics module is used to take the extracted key features as input, analyze the correlation parameters through time series analysis of long short-term memory network, and establish a blockage cause knowledge base and classify and label blockage types. The risk warning module is used to construct a congestion risk classification model based on the congestion cause library and machine learning algorithm, and generate a three-level warning. The blockage clearing control and execution module is used to activate the corresponding air cannon based on the matching result of the level corresponding to the three-level warning and the visual positioning, dynamically adjust the injection parameters, and execute the blockage clearing operation according to the optimized timing sequence.

Images