CN113903007A - Intelligent scene analysis system for water conservancy industry - Google Patents
Intelligent scene analysis system for water conservancy industry Download PDFInfo
- Publication number
- CN113903007A CN113903007A CN202111504415.9A CN202111504415A CN113903007A CN 113903007 A CN113903007 A CN 113903007A CN 202111504415 A CN202111504415 A CN 202111504415A CN 113903007 A CN113903007 A CN 113903007A
- Authority
- CN
- China
- Prior art keywords
- algorithm
- intelligent
- image
- water
- picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
An intelligent scenario analysis system for the water conservancy industry, comprising: the intelligent algorithm management unit is used for butting the data terminals and classifying and managing the image data transmitted by the data terminals; the intelligent algorithm management unit is also used for connecting a user side and outputting data; the intelligent algorithm management unit is also connected to the intelligent algorithm training unit and acquires an identifier algorithm from the intelligent algorithm training unit, and the identifier algorithm processes and analyzes the image data; the intelligent algorithm reasoning unit comprises a localization center computing module; and the intelligent algorithm training unit is provided with an AI intelligent video analysis management module for generating an identifier algorithm. The system can dynamically supervise and early-warning and analyze the hydraulic engineering in real time, and realize intelligent patrol and supervision of the hydraulic engineering; the method can be used for analyzing according to a specific recognizer algorithm as required, and has high analysis efficiency and high accuracy of an analysis result.
Description
Technical Field
The invention belongs to the technical field of scene analysis in the water conservancy industry, and particularly relates to an intelligent scene analysis system for the water conservancy industry.
Background
In recent years, the climate change is aggravated, the precipitation of the south of China, most parts of south China, the south of the southwest and the like is obviously less than the same period of the year, the precipitation of part of areas is more than six times less than the same period of the year, the water level of small and medium-sized reservoirs in part of areas is close to or lower than the dead water level, water supply shortage conditions occur in part of cities and towns, meanwhile, the precipitation of the middle area and the north area is obviously increased more than the same period of the year, and even natural disasters such as flood are caused.
For some large urban areas, the water level situation of the current river channel is accurately analyzed under the conditions of severe rainfall change and insufficient water resource supply, and urban networking water supply allocation and cross-regional water diversion scheduling can be effectively supported to guarantee water supply safety. Meanwhile, river monitoring, reservoir monitoring, water supply network monitoring, pollution discharge monitoring and the like are required to be involved in the water conservancy industry, the work is generally carried out by manual inspection, the workload of the manual inspection is large, and accidents are easy to happen in some dangerous areas.
In the prior art, some simple analysis devices based on camera images for the water conservancy industry exist, but most of the analysis devices are simple in structure, lag in functions and poor in analysis precision, and are difficult to meet the higher and higher water conservancy scene analysis requirements.
Therefore, aiming at the problems in the prior art, the invention further improves the intelligent scene analysis system used in the water conservancy industry.
Disclosure of Invention
To the not enough among the prior art above, the application provides an intelligent scene analytic system for water conservancy industry, the camera in the concrete scene of cooperation, unmanned aerial vehicle, unmanned vehicle data acquisition, adopt intelligent unit of modularization overall formula, can carry out real-time dynamic's supervision and early warning analysis to water conservancy projects such as reservoir, dyke, sea pond, river course, sluice, pump station, realize water conservancy project intelligence inspection and supervision, and can discover the violation event very first time, report and handle, maintain the healthy life in each waters, realize economic society sustainable development.
The technical scheme in the application is realized through the following technical scheme.
An intelligent scenario analysis system for the water conservancy industry, comprising: the intelligent algorithm management unit is used for butting the data terminals and classifying and managing the image data transmitted by the data terminals; the intelligent algorithm management unit is also used for connecting a user side and outputting data; the intelligent algorithm management unit is also connected to the intelligent algorithm training unit and acquires an identifier algorithm from the intelligent algorithm training unit, and the identifier algorithm processes and analyzes the image data; the intelligent algorithm reasoning unit comprises a localization center computing module, and the localization center computing module is used for distributing computing resources and executing intelligent task scheduling service and intelligent analysis basic service; the system is also used for calling storage resources to perform data comparison and algorithm management services; the intelligent algorithm training unit is internally provided with an AI intelligent video analysis management module, the AI intelligent video analysis management module is used for generating a trained identifier algorithm, and the identifier algorithm comprises: at least one of a floater identification algorithm, an illegal wading activity identification algorithm, an illegal drainage identification algorithm, a perimeter identification algorithm, a water gauge identification algorithm, a water body area identification algorithm and a safety helmet wearing identification algorithm.
In this application, the data end can be for the camera in the concrete water conservancy scene for the fixed point shoots image, video, also can be for loading unmanned aerial vehicle or the unmanned vehicles of camera device and communication device, can carry out automatic patrol according to the instruction program and shoot, also can be for the camera device that patrol personnel took, and picture, image data that these shoot are passed into to intelligent algorithm administrative unit through the data end.
After the data are obtained, the intelligent algorithm management unit classifies and manages the image data, for example, the image data are classified into reservoir scene image data, river scene image data, personnel work area image data, gate dam area image data and the like according to region classification, then a proper identification sub algorithm is selected from the intelligent algorithm training unit according to needs, the image data are analyzed and processed, and a processing result is fed back to a connection user side.
The user side can be a user computer, a smart phone and other equipment of the terminal, can check image data, analysis results and the like, and can customize required analysis services to be executed by the intelligent algorithm management unit.
The localization center computing module may be a local server for providing hardware and computational support, as well as for storing data. In an optimal scheme, the localization center computing module can be used in cooperation with the front-end node settlement module, and a center + edge computing mode is adopted, so that the data analysis efficiency is improved, and the computing power is saved. The front-end node settlement module can be an AI camera with the functions of preliminary analysis, classification and calculation, and can perform preliminary data processing on the shot images.
The intelligent algorithm training unit is a customized upgradable module and is used for creating, verifying, training and issuing an algorithm model, and producing an identifier algorithm with corresponding functions after issuing for the intelligent algorithm management unit to select and use.
In a preferred embodiment, the float identification algorithm is configured to perform the following operations: s11: analyzing the image or picture collected by the data end, and detecting and positioning the floater on the image or picture; s12: performing pixel-level segmentation on the image and the picture, and judging the detection result of the floater by combining the collision detection of the floater and the water body; s13: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the illegal wading activity identification algorithm is configured to: s21, analyzing the image or picture collected by the data terminal, and detecting and positioning the human body and/or the fishing rod for the image or picture; s22: performing pixel-level water body segmentation on the image and the picture, and judging whether the swimming occurs or not through collision between the water body segmentation and a human body region; and/or: performing pixel-level water body segmentation on the image and the picture, and judging whether fishing is performed or not through collision between a fishing rod area and a human body area; s23: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the violation drain identification algorithm is configured to perform the following operations: s31: analyzing the image or picture acquired by the data terminal, and detecting and positioning the sewage outlet on the image or picture; s32: performing pixel-level water body segmentation on the image and the picture, and judging whether sewage is discharged or not through the water body segmentation and the area collision detection of a sewage discharge outlet; s33: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the perimeter identification algorithm is configured to perform the following operations: s41: analyzing the image or picture collected by the data end, and detecting and positioning the human body on the image or picture; s42: the method comprises the following steps of (1) delimiting forbidden zone boundaries of images and pictures, and judging whether a human body is overlapped with the forbidden zone or not through detecting the collision of the forbidden zone boundaries and the human body; s43: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the water gauge identification algorithm is configured to perform the following operations: s51: analyzing the image or picture collected by the data end, and carrying out intelligent identification of the water gauge on the image or picture; s52: analyzing water level lines of the water gauges in the images and the pictures, comparing the water level values with original calibration images, and calculating water level values; s53: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the water area identification algorithm is configured to perform the following operations: s61: analyzing the image or picture acquired by the data end, and carrying out pixel-level segmentation on the image or picture in the image or picture; s62: comparing the image and the picture with the original calibrated image, and judging the boundary of the water area of the water body; s63: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, the headgear wearing identification algorithm is configured to: s71: analyzing the images or pictures acquired by the data terminal, detecting and positioning the images or pictures in the images or pictures to detect the wearing condition of the safety helmet on the head of the human body; s72: comparing the image and the picture with the original calibrated image, and judging the boundary of the wearing area of the safety helmet; s73: carrying out collision detection on the wearing condition of the safety helmet and the wearing area of the safety helmet, and judging whether a human body in the wearing area of the safety helmet wears the safety helmet or not; s74: and feeding back the detection result to the intelligent algorithm management unit.
In a preferred embodiment, in the identification sub-algorithm, the processing of the pictures and images may adopt one or more of a multi-level resolution pixel level scene segmentation algorithm, a two-level fine positioning few-sample object detection algorithm, and a big data driven difficult/easy sample adaptive human body detection algorithm.
In a preferred embodiment, in the present application, the intelligent algorithm inference unit performs statistics on the quality of the recognizer algorithms running therein, and feeds the statistics back to the intelligent algorithm training unit for optimization training.
Compared with the prior art, the method has the following beneficial effects: the intelligent scene analysis system for the water conservancy industry is provided, is matched with a camera, an unmanned aerial vehicle and an unmanned vehicle in a specific scene for data acquisition, adopts a modularized overall intelligent unit, and can perform real-time dynamic supervision and early warning analysis on water conservancy projects such as reservoirs, embankments, ponds, riverways, water gates, pump stations and the like, so as to realize intelligent patrol and supervision of the water conservancy projects; the method can be used for analyzing according to a specific recognizer algorithm as required, and has high analysis efficiency and high accuracy of an analysis result.
Drawings
Fig. 1 is a design flowchart of an intelligent scene analysis system according to the present application.
Fig. 2 is a unit relationship diagram of the intelligent scene analysis system in the present application.
Fig. 3 is a flow chart of the float detection process in the present application.
Fig. 4 is a flow of illegal wading activity detection processing in the present application.
Fig. 5 is a flow chart of the illegal water discharge detection processing in the present application.
Fig. 6 is a flowchart of perimeter intrusion detection processing in the present application.
Fig. 7 is a flowchart of the water gauge water level detection process in the present application.
Fig. 8 is a flow chart of a water area detection process according to the present application.
Fig. 9 is a flowchart of the helmet wearing detection process according to the present application.
FIG. 10 is a schematic diagram of a multi-level resolution pixel level segmentation algorithm in the present application.
Fig. 11 is a schematic diagram of a multi-level pixel-level water body segmentation depth algorithm in the present application.
Fig. 12 is a schematic diagram of a target rough location detection algorithm in the present application.
Fig. 13 is a schematic diagram of a target fine positioning detection algorithm in the present application.
Fig. 14 is a schematic diagram of an adaptive hard/easy sample detection algorithm in the present application.
FIG. 15 is a schematic diagram of a training optimization closed-loop algorithm in the present application.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings and the detailed description.
An intelligent scene analysis system for water conservancy industry in this application includes: the intelligent algorithm management unit is used for butting the data terminals and classifying and managing the image data transmitted by the data terminals; the intelligent algorithm management unit is also used for connecting a user side and outputting data; the intelligent algorithm management unit is also connected to the intelligent algorithm training unit and acquires an identifier algorithm from the intelligent algorithm training unit, and the identifier algorithm processes and analyzes the image data; the intelligent algorithm reasoning unit comprises a localization center computing module, and the localization center computing module is used for distributing computing resources and executing intelligent task scheduling service and intelligent analysis basic service; the system is also used for calling storage resources to perform data comparison and algorithm management services; the intelligent algorithm training unit is internally provided with an AI intelligent video analysis management module, the AI intelligent video analysis management module is used for generating a trained identifier algorithm, and the identifier algorithm comprises: at least one of a floater identification algorithm, an illegal wading activity identification algorithm, an illegal drainage identification algorithm, a perimeter identification algorithm, a water gauge identification algorithm, a water body area identification algorithm and a safety helmet wearing identification algorithm.
In this application, the data end can be for the camera in the concrete water conservancy scene for the fixed point shoots image, video, also can be for loading unmanned aerial vehicle or the unmanned vehicles of camera device and communication device, can carry out automatic patrol according to the instruction program and shoot, also can be for the camera device that patrol personnel took, and picture, image data that these shoot are passed into to intelligent algorithm administrative unit through the data end.
After the data are obtained, the intelligent algorithm management unit classifies and manages the image data, for example, the image data are classified into reservoir scene image data, river scene image data, personnel work area image data, gate dam area image data and the like according to region classification, then a proper identification sub algorithm is selected from the intelligent algorithm training unit according to needs, the image data are analyzed and processed, and a processing result is fed back to a connection user side.
The user side can be a user computer, a smart phone and other equipment of the terminal, can check image data, analysis results and the like, and can customize required analysis services to be executed by the intelligent algorithm management unit.
The localization center computing module may be a local server for providing hardware and computational support, as well as for storing data. In an optimal scheme, the localization center computing module can be used in cooperation with the front-end node settlement module, and a center + edge computing mode is adopted, so that the data analysis efficiency is improved, and the computing power is saved. The front-end node settlement module can be an AI camera with the functions of preliminary analysis, classification and calculation, and can perform preliminary data processing on the shot images.
The intelligent algorithm training unit is a customized upgradable module and is used for creating, verifying, training and issuing an algorithm model, and producing an identifier algorithm with corresponding functions after issuing for the intelligent algorithm management unit to select and use.
Referring to fig. 1, it can be seen that the core of the intelligent scene analysis system for the water conservancy industry in the application is a customized intelligent image analysis technology based on customer requirements. Specifically, an AI intelligent video analysis management module (i.e., the AI intelligent visual analysis management platform in fig. 1) completes the collection and data processing of the requirements according to the application requirements of the water conservancy service scene, trains and generates an intelligent video analysis algorithm through the AI intelligent video analysis management platform, realizes local deployment application and statistical evaluation, and performs targeted optimization and promotion on the algorithm with too low accuracy according to the statistical condition, so that the algorithm reaching the standard can be further popularized and applied, and the upper system is used for the user to select and use the function.
As can be seen from fig. 2, in the present application, the intelligent algorithm management unit in the present application may be docked to each level of water conservancy network video monitoring platform, and is configured to acquire video data in the industry and feed back the analyzed early warning data to the platform; meanwhile, the system can be connected to all levels of water conservancy service platforms in a butt joint mode and used for interaction and collection requirements.
The recognizer algorithm in the present application is specifically seen below.
(1) Floater identification algorithm: the float identification algorithm is configured to perform the following operations: s11: analyzing the image or picture collected by the data end, and detecting and positioning the floater on the image or picture; s12: performing pixel-level segmentation on the image and the picture, and judging the detection result of the floater by combining the collision detection of the floater and the water body; s13: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen in fig. 3, the system receives a service request sent from a user, and the intelligent algorithm management unit calls the floating object identification algorithm to decode the image: the method comprises the steps of positioning floaters in an image through an algorithm, simultaneously segmenting a water body, filtering objects which are not on the water through collision detection of the floaters and the water body, and outputting the result if the floaters are left on the water.
(2) An illegal wading activity recognition algorithm: the illegal wading activity recognition algorithm is used for executing the following operations: s21, analyzing the image or picture collected by the data terminal, and detecting and positioning the human body and/or the fishing rod for the image or picture; s22: performing pixel-level water body segmentation on the image and the picture, and judging whether the swimming occurs or not through collision between the water body segmentation and a human body region; and/or: performing pixel-level water body segmentation on the image and the picture, and judging whether fishing is performed or not through collision between a fishing rod area and a human body area; s23: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 4, the system receives a service request sent by a user, analyzes an image acquired by a front end, performs main human body detection on the images one by one, performs water body segmentation and fishing rod detection positioning simultaneously, judges whether to swim or not through collision between the water body segmentation and a human body region, judges whether to fish or not through collision between the fishing rod region and the human body region, and feeds a detection result back to the intelligent algorithm management unit.
(3) And (3) illegal drainage identification algorithm: the illegal water drainage identification algorithm is used for executing the following operations: s31: analyzing the image or picture acquired by the data terminal, and detecting and positioning the sewage outlet on the image or picture; s32: performing pixel-level water body segmentation on the image and the picture, and judging whether sewage is discharged or not through the water body segmentation and the area collision detection of a sewage discharge outlet; s33: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 5, the system receives a service request sent by a user side, analyzes an image collected by a front end, detects the positions of the sewage outlets one by one on the image, simultaneously identifies whether water is discharged, judges a drainage detection result, and feeds the detection result back to the intelligent algorithm management unit.
(4) And (3) a perimeter identification algorithm: the perimeter identification algorithm is configured to perform the following operations: s41: analyzing the image or picture collected by the data end, and detecting and positioning the human body on the image or picture; s42: the method comprises the following steps of (1) delimiting forbidden zone boundaries of images and pictures, and judging whether a human body is overlapped with the forbidden zone or not through detecting the collision of the forbidden zone boundaries and the human body; s43: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 6, the system receives a service request sent by a user, analyzes an image acquired by a front end, detects the occurrence of a human body one by one for the image, determines whether the position where the human body appears overlaps with a forbidden zone according to a defined forbidden zone boundary, performs zone collision detection, thereby determining whether the human body breaks into the forbidden zone, and feeds back a detection result to the intelligent algorithm management unit.
(5) And (3) water gauge identification algorithm: the water gauge identification algorithm is used for executing the following operations: s51: analyzing the image or picture collected by the data end, and carrying out intelligent identification of the water gauge on the image or picture; s52: analyzing water level lines of the water gauges in the images and the pictures, comparing the water level values with original calibration images, and calculating water level values; s53: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 7, the system receives a service request sent by a user, calls a front-end water gauge camera, intelligently identifies the water gauge, analyzes images acquired by the front end, performs water level line analysis on the images one by one, compares the images with original calibration images, calculates a water level value, and feeds back a final detection result to the intelligent algorithm management unit after multiple average measurements and calculations.
(6) A water body and water area identification algorithm: the water body area identification algorithm is used for executing the following operations: s61: analyzing the image or picture acquired by the data end, and carrying out pixel-level segmentation on the image or picture in the image or picture; s62: comparing the image and the picture with the original calibrated image, and judging the boundary of the water area of the water body; s63: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 8, the system receives a service request sent by a user, analyzes an image acquired by a front end, performs pixel-level segmentation on the images of the water body one by one, compares the segmented images with an original calibrated image, determines the boundary of the water body, and feeds back a detection result to the intelligent algorithm management unit.
(9) The safety helmet wearing identification algorithm comprises the following steps: the headgear wearing identification algorithm is configured to perform the following operations: s71: analyzing the images or pictures acquired by the data terminal, detecting and positioning the images or pictures in the images or pictures to detect the wearing condition of the safety helmet on the head of the human body; s72: comparing the image and the picture with the original calibrated image, and judging the boundary of the wearing area of the safety helmet; s73: carrying out collision detection on the wearing condition of the safety helmet and the wearing area of the safety helmet, and judging whether a human body in the wearing area of the safety helmet wears the safety helmet or not; s74: and feeding back the detection result to the intelligent algorithm management unit.
Specifically, as can be seen from fig. 9, the system receives a service request sent by a user, analyzes an image acquired by a front end, detects the occurrence of a human body in a single image, positions the head of the human body entering a detection area, detects a safety helmet, judges whether the human body wears the safety helmet or not through the collision between the head area of the human body and the safety helmet, and feeds back a detection result to the intelligent algorithm management unit.
In the present application, in the above identification sub-algorithms, the processing of the picture and the image may adopt one or more of a multi-level resolution pixel level scene segmentation algorithm, a two-level fine positioning few-sample target detection algorithm, and a big data driving difficult/easy-sample adaptive human body detection algorithm.
Specifically, the multi-level resolution pixel level scene segmentation algorithm is described with reference to fig. 10 and fig. 11. The identification sub-algorithm of the system relates to the detection of targets on a wide range of water bodies, and the accurate pixel-level water body segmentation algorithm is required to enable the water body to be accurately segmented in various complex scenes. The pixel-level water body segmentation algorithm analyzes segmented information under various levels of resolution (such as 4 levels of resolution), simultaneously acquires rich pixel and region information, and simultaneously exchanges information analysis information under information analysis paths of various levels of resolution by adopting a multi-resolution information collection module (shown in figure 10), thereby greatly enhancing the expression of the pixel and region information under different paths and achieving accurate pixel segmentation. On the other hand, as shown in fig. 11, by the deep network structure of 76 layers, richer division information is extracted. The accurate water body pixel segmentation is completed, and the method is effectively used for floater detection, water body water area detection, illegal wading activity detection and the like.
Specifically, the two-stage fine positioning few-sample target detection algorithm is described with reference to fig. 12 and 13.
The floating object detection, the illegal wading activity detection, the illegal drainage detection, the safe wearing detection and the like relate to a fine target detection algorithm (the target is probably smaller under a long distance), and the characteristics of few samples exist. According to the characteristics, a two-stage gradually-refined positioning detection depth model structure is designed, so that the method is suitable for more accurate target detection under the condition of few samples, and the following embodiment is provided.
1. The first-level depth model structure adopts a deep trunk depth model structure with 50 layers to extract image features, and constructs multi-scale features through a layer-by-layer up-sampling module, wherein each scale feature is responsible for detecting targets with different scales, the features with high resolution (large scale) detect small targets, and the features with low resolution (small scale) detect large targets. Different detection targets roughly estimate target areas through a first-level depth model structure.
2. And the second-level depth model structure estimates a rough target region according to the first-level structure, further extracts target features in the region, and performs fine target region estimation and confidence degree estimation through a depth model of 5 layers of linear regression layers. On the other hand, in the case of fewer positive samples, the second-level depth model is optimized by adding a large number of samples, and the detection accuracy under the condition of fewer samples is optimized.
Specifically, a big data driven hard/easy sample adaptive human body detection algorithm is described with reference to fig. 14.
Water conservancy video surveillance cameras are distributed in various water conservancy scenes, and have many complex factors influencing detection, such as angles, heights and distances of the cameras, illumination brightness change of the environment, complex backgrounds (backgrounds of various plants, ships, water bodies and the like) and the like. Therefore, the following two aspects are adopted for solving the problems: (1) and mass human body labeling data. (2) An adaptive hard/easy sample detection algorithm.
A large number of scenes which are difficult to detect probably exist in water conservancy data distribution, and a part of scenes which are easy to detect are designed with a self-adaptive difficult/easy sample detection algorithm, so that (1) the scenes which are easy to detect are directly inferred and detected through a small network, and the time is saved; (2) and 3, reasoning and detecting the scene difficult to detect through a large network, and ensuring the precision of the scene difficult to detect.
The core of the self-adaptive hard/easy sample detection algorithm is a hard/easy sample decision module, and images under complex factors are subjected to an ultra-deep detection network through the decision of the hard/easy sample decision module, so that the detection precision of the hard samples under the complex factors is ensured. On the other hand, the self-adaptive human body detection depth model has higher detection precision under various scenes through a data-driven training mode by training a network through mass human body image data, such as 50 thousands of image data including image data under various scenes such as night, day, riverway, reservoir, water conservancy project and the like.
In addition, in the present application, the intelligent algorithm inference unit performs statistics on the quality of the recognizer algorithm running therein, and feeds the statistics back to the intelligent algorithm training unit for optimization training, so as to form an effective closed loop, specifically, see with reference to fig. 15: by means of actual application of the intelligent algorithm, application effects of different algorithms on a plurality of scenes are summarized, accuracy statistics is carried out on recognition results, the statistical results are fed back to the algorithm model, algorithm analysis promotion and optimization work is carried out, algorithm optimization is carried out on the algorithm model in a targeted mode, deep learning is further carried out, the optimized algorithm is integrated and applied, accordingly analysis capability of the intelligent algorithm is continuously promoted, complementary circulation of intelligent algorithm application and optimization is formed, and effective closed loop of training, calculation and optimization is achieved.
In addition, in this application, in the system of this application, still be equipped with intelligent video analysis and support and evaluation unit, it is specific: the intelligent video analysis supporting and evaluating unit integrates access, generation, training, optimization and evaluation intelligent analysis algorithms, so that the system is more in line with the requirements of water conservancy projects such as reservoirs, dikes, sea ponds, riverways, water gates and pump stations on real-time dynamic environment supervision, and the water conservancy automation and intelligent management level is further improved.
The system in the application realizes the unified management of multiple algorithms, so that each system can call intelligent analysis algorithms for specific targets and specific behaviors according to requirements. Meanwhile, the integrated intelligent analysis algorithm is subjected to analysis basic services and statistical evaluation processing, the analysis services are managed in a unified mode, interfaces of various services are opened, a service interface with water conservancy intelligent video analysis characteristics is provided, a user can flexibly arrange the services according to the requirements of the water conservancy industry, and the running state and the quality of the services are well monitored and controlled. Meanwhile, in order to improve the phenomena of false alarm and missing report in the intelligent video analysis process, algorithm optimization work is carried out, from the confirmation of the algorithm optimization direction to the targeted collection of samples for training, and the system realizes the improvement of the algorithm analysis precision through a circulating training mode.
The above, the application provides an intelligent scene analysis system for the water conservancy industry, which is matched with data acquisition of a camera, an unmanned aerial vehicle and an unmanned vehicle in a specific scene, adopts a modularized integrated intelligent unit, and can perform real-time dynamic supervision and early warning analysis on water conservancy projects such as reservoirs, embankments, ponds, riverways, water gates, pump stations and the like, so as to realize intelligent patrol and supervision of the water conservancy projects; the method can be used for analyzing according to a specific recognizer algorithm as required, and has high analysis efficiency and high accuracy of an analysis result.
The scope of the present invention includes, but is not limited to, the above embodiments, and the present invention is defined by the appended claims, and any alterations, modifications, and improvements that may occur to those skilled in the art are all within the scope of the present invention.
Claims (10)
1. An intelligent scene analysis system for water conservancy industry, comprising:
the intelligent algorithm management unit is used for butting the data terminals and classifying and managing the image data transmitted by the data terminals; the intelligent algorithm management unit is also used for connecting a user side and outputting data; the intelligent algorithm management unit is also connected to the intelligent algorithm training unit and acquires an identifier algorithm from the intelligent algorithm training unit, and the identifier algorithm processes and analyzes the image data;
the intelligent algorithm reasoning unit comprises a localization center computing module, and the localization center computing module is used for distributing computing resources and executing intelligent task scheduling service and intelligent analysis basic service; the system is also used for calling storage resources to perform data comparison and algorithm management services;
the intelligent algorithm training unit is internally provided with an AI intelligent video analysis management module, the AI intelligent video analysis management module is used for generating a trained identifier algorithm, and the identifier algorithm comprises: at least one of a floater identification algorithm, an illegal wading activity identification algorithm, an illegal drainage identification algorithm, a perimeter identification algorithm, a water gauge identification algorithm, a water body area identification algorithm and a safety helmet wearing identification algorithm.
2. An intelligent scenario analysis system for the water conservancy industry according to claim 1, wherein the float identification algorithm is configured to perform the following operations:
s11: analyzing the image or picture collected by the data end, and detecting and positioning the floater on the image or picture;
s12: performing pixel-level segmentation on the image and the picture, and judging the detection result of the floater by combining the collision detection of the floater and the water body;
s13: and feeding back the detection result to the intelligent algorithm management unit.
3. The intelligent scene analysis system for the water conservancy industry according to claim 1, wherein the illegal wading activity recognition algorithm is used for performing the following operations:
s21, analyzing the image or picture collected by the data terminal, and detecting and positioning the human body and/or the fishing rod for the image or picture;
s22: performing pixel-level water body segmentation on the image and the picture, and judging whether the swimming occurs or not through collision between the water body segmentation and a human body region; and/or: performing pixel-level water body segmentation on the image and the picture, and judging whether fishing is performed or not through collision between a fishing rod area and a human body area;
s23: and feeding back the detection result to the intelligent algorithm management unit.
4. The intelligent scenario analysis system for water conservancy industry of claim 1, wherein the illegal water draining identification algorithm is configured to perform the following operations:
s31: analyzing the image or picture acquired by the data terminal, and detecting and positioning the sewage outlet on the image or picture;
s32: performing pixel-level water body segmentation on the image and the picture, and judging whether sewage is discharged or not through the water body segmentation and the area collision detection of a sewage discharge outlet;
s33: and feeding back the detection result to the intelligent algorithm management unit.
5. The intelligent scenario analysis system of claim 1, wherein the perimeter recognition algorithm is configured to perform the following operations:
s41: analyzing the image or picture collected by the data end, and detecting and positioning the human body on the image or picture;
s42: the method comprises the following steps of (1) delimiting forbidden zone boundaries of images and pictures, and judging whether a human body is overlapped with the forbidden zone or not through detecting the collision of the forbidden zone boundaries and the human body;
s43: and feeding back the detection result to the intelligent algorithm management unit.
6. The intelligent scene analysis system for the water conservancy industry according to claim 1, wherein the water gauge identification algorithm is used for performing the following operations:
s51: analyzing the image or picture collected by the data end, and carrying out intelligent identification of the water gauge on the image or picture;
s52: analyzing water level lines of the water gauges in the images and the pictures, comparing the water level values with original calibration images, and calculating water level values;
s53: and feeding back the detection result to the intelligent algorithm management unit.
7. The intelligent scene analysis system for the water conservancy industry according to claim 1, wherein the water body and water area identification algorithm is configured to perform the following operations:
s61: analyzing the image or picture acquired by the data end, and carrying out pixel-level segmentation on the image or picture in the image or picture;
s62: comparing the image and the picture with the original calibrated image, and judging the boundary of the water area of the water body;
s63: and feeding back the detection result to the intelligent algorithm management unit.
8. The intelligent scene analysis system for the water conservancy industry according to claim 1, wherein the helmet wearing identification algorithm is used for performing the following operations:
s71: analyzing the images or pictures acquired by the data terminal, detecting and positioning the images or pictures in the images or pictures to detect the wearing condition of the safety helmet on the head of the human body;
s72: comparing the image and the picture with the original calibrated image, and judging the boundary of the wearing area of the safety helmet;
s73: carrying out collision detection on the wearing condition of the safety helmet and the wearing area of the safety helmet, and judging whether a human body in the wearing area of the safety helmet wears the safety helmet or not;
s74: and feeding back the detection result to the intelligent algorithm management unit.
9. The system according to any one of claims 1 to 8, wherein the recognition sub-algorithm is one or more of a multi-level resolution pixel level scene segmentation algorithm, a two-level fine-positioning few-sample object detection algorithm, and a big data-driven hard/easy-sample adaptive human body detection algorithm.
10. The intelligent scene analysis system for the water conservancy industry according to claim 9, wherein the intelligent algorithm reasoning unit counts the quality of the recognizer algorithm running in the intelligent algorithm reasoning unit and feeds the quality back to the intelligent algorithm training unit for optimization training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111504415.9A CN113903007A (en) | 2021-12-10 | 2021-12-10 | Intelligent scene analysis system for water conservancy industry |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111504415.9A CN113903007A (en) | 2021-12-10 | 2021-12-10 | Intelligent scene analysis system for water conservancy industry |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113903007A true CN113903007A (en) | 2022-01-07 |
Family
ID=79025568
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111504415.9A Pending CN113903007A (en) | 2021-12-10 | 2021-12-10 | Intelligent scene analysis system for water conservancy industry |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113903007A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117714484A (en) * | 2023-11-15 | 2024-03-15 | 杭州中汇通航航空科技有限公司 | Unmanned aerial vehicle flight report management platform |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101194167B1 (en) * | 2012-05-22 | 2012-10-24 | (주)버들시스템 | Trespassing detection and video recoding system for small scale water supply establishment |
| KR101866239B1 (en) * | 2017-05-30 | 2018-06-12 | (주)이피에스이앤이 | Method for Monitoring Water Quality Environment Using Drone |
| CN108985274A (en) * | 2018-08-20 | 2018-12-11 | 上海磐波智能科技有限公司 | Water surface method for recognizing impurities |
| CN109405808A (en) * | 2018-10-19 | 2019-03-01 | 天津英田视讯科技有限公司 | A kind of hydrological monitoring spherical camera |
| CN110084210A (en) * | 2019-04-30 | 2019-08-02 | 电子科技大学 | The multiple dimensioned Ship Detection of SAR image based on attention pyramid network |
| CN110344495A (en) * | 2019-06-20 | 2019-10-18 | 中水淮河规划设计研究有限公司 | Pumping plant video identification dirt intelligence is removed contamination system and intelligent cleaning method |
| CN110427836A (en) * | 2019-07-11 | 2019-11-08 | 重庆市地理信息和遥感应用中心(重庆市测绘产品质量检验测试中心) | A kind of high-resolution remote sensing image Clean water withdraw method based on multi_dimension optimization |
| CN110781775A (en) * | 2019-10-10 | 2020-02-11 | 武汉大学 | Remote sensing image water body information accurate segmentation method supported by multi-scale features |
| CN112418128A (en) * | 2020-11-30 | 2021-02-26 | 重庆市生态环境大数据应用中心 | Surface water monitoring and management system and method |
| CN112532921A (en) * | 2020-10-28 | 2021-03-19 | 深圳英飞拓科技股份有限公司 | Water conservancy system intelligent monitoring implementation method and system |
| CN112766274A (en) * | 2021-02-01 | 2021-05-07 | 长沙市盛唐科技有限公司 | Water gauge image water level automatic reading method and system based on Mask RCNN algorithm |
| CN113536836A (en) * | 2020-04-15 | 2021-10-22 | 宁波弘泰水利信息科技有限公司 | Method for monitoring river and lake water area encroachment based on unmanned aerial vehicle remote sensing technology |
-
2021
- 2021-12-10 CN CN202111504415.9A patent/CN113903007A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101194167B1 (en) * | 2012-05-22 | 2012-10-24 | (주)버들시스템 | Trespassing detection and video recoding system for small scale water supply establishment |
| KR101866239B1 (en) * | 2017-05-30 | 2018-06-12 | (주)이피에스이앤이 | Method for Monitoring Water Quality Environment Using Drone |
| CN108985274A (en) * | 2018-08-20 | 2018-12-11 | 上海磐波智能科技有限公司 | Water surface method for recognizing impurities |
| CN109405808A (en) * | 2018-10-19 | 2019-03-01 | 天津英田视讯科技有限公司 | A kind of hydrological monitoring spherical camera |
| CN110084210A (en) * | 2019-04-30 | 2019-08-02 | 电子科技大学 | The multiple dimensioned Ship Detection of SAR image based on attention pyramid network |
| CN110344495A (en) * | 2019-06-20 | 2019-10-18 | 中水淮河规划设计研究有限公司 | Pumping plant video identification dirt intelligence is removed contamination system and intelligent cleaning method |
| CN110427836A (en) * | 2019-07-11 | 2019-11-08 | 重庆市地理信息和遥感应用中心(重庆市测绘产品质量检验测试中心) | A kind of high-resolution remote sensing image Clean water withdraw method based on multi_dimension optimization |
| CN110781775A (en) * | 2019-10-10 | 2020-02-11 | 武汉大学 | Remote sensing image water body information accurate segmentation method supported by multi-scale features |
| CN113536836A (en) * | 2020-04-15 | 2021-10-22 | 宁波弘泰水利信息科技有限公司 | Method for monitoring river and lake water area encroachment based on unmanned aerial vehicle remote sensing technology |
| CN112532921A (en) * | 2020-10-28 | 2021-03-19 | 深圳英飞拓科技股份有限公司 | Water conservancy system intelligent monitoring implementation method and system |
| CN112418128A (en) * | 2020-11-30 | 2021-02-26 | 重庆市生态环境大数据应用中心 | Surface water monitoring and management system and method |
| CN112766274A (en) * | 2021-02-01 | 2021-05-07 | 长沙市盛唐科技有限公司 | Water gauge image water level automatic reading method and system based on Mask RCNN algorithm |
Non-Patent Citations (4)
| Title |
|---|
| COLIN VAN LIESHOUT 等: "Automated River Plastic Monitoring Using Deep Learning and Cameras", 《EARTH AND SPACE SCIENCE》 * |
| KUEI-CHUNG CHANG 等: "Design of River Monitoring System Based on Video Recognition and Deep Learning Technologies", 《2020 IEEE INTERNATIONAL CONFERENCE ON CONSUMER ELECTRONICS》 * |
| 程诚 等: "智慧视频识别在水利信息化中的应用", 《四川水利》 * |
| 雷佳明 等: "人工智能图像识别在水库智能监控方面的应用", 《中国水利学会2019学术年会论文集》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117714484A (en) * | 2023-11-15 | 2024-03-15 | 杭州中汇通航航空科技有限公司 | Unmanned aerial vehicle flight report management platform |
| CN117714484B (en) * | 2023-11-15 | 2024-06-18 | 杭州中汇通航航空科技有限公司 | Unmanned aerial vehicle flight report management platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101996410B (en) | Method and system of detecting moving object under dynamic background | |
| CN101957997B (en) | Regional average value kernel density estimation-based moving target detecting method in dynamic scene | |
| CN104660993B (en) | Maritime affairs intelligent control method and system based on AIS and CCTV | |
| CN101188745B (en) | Intelligent drowning video monitoring system and method for natatorium | |
| CN115603466B (en) | Ship shore power system based on artificial intelligence visual identification | |
| CN114049477B (en) | Fish passing fishway system and dynamic identification and tracking method for fish quantity and fish type | |
| CN116152748B (en) | River and lake supervision method and system based on blue algae identification | |
| CN109255350A (en) | A kind of new energy detection method of license plate based on video monitoring | |
| CN111582182B (en) | Ship name recognition method, system, computer equipment and storage medium | |
| CN113903007A (en) | Intelligent scene analysis system for water conservancy industry | |
| CN112802064A (en) | Visual identification system for fishway of hydro-junction | |
| WO2021146617A1 (en) | Image based aquatic alert system | |
| CN111680610A (en) | Construction scene abnormity monitoring method and device | |
| CN101750061B (en) | Method and device for target flight path detection/course prediction | |
| CN109521021A (en) | A kind of nuclear power plant containment shell appearance inspecting system and method | |
| CN109146923A (en) | The processing method and system of disconnected frame are lost in a kind of target following | |
| CN115171054A (en) | Fishing administration patrol method and system | |
| CN113869683A (en) | Nuclear power plant cold source safety early warning method and device, computer equipment and system | |
| CN114220010A (en) | Intelligent image processing method suitable for intelligent analysis system in water conservancy industry | |
| CN101567088B (en) | Method and device for detecting moving object | |
| US20240153109A1 (en) | Image based tracking system | |
| CN116246204A (en) | Monitoring method, device and system for sand production behavior and storage medium | |
| CN110210436A (en) | A kind of vehicle-mounted camera line walking image-recognizing method | |
| CN113542697B (en) | Adaptive monitoring method of sand production ship adaptive monitoring system based on deep learning | |
| CN214475140U (en) | Visual identification system for fishway of hydro-junction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220107 |
|
| RJ01 | Rejection of invention patent application after publication |