CN110031597A - A kind of biological water monitoring method - Google Patents
A kind of biological water monitoring method Download PDFInfo
- Publication number
- CN110031597A CN110031597A CN201910316179.4A CN201910316179A CN110031597A CN 110031597 A CN110031597 A CN 110031597A CN 201910316179 A CN201910316179 A CN 201910316179A CN 110031597 A CN110031597 A CN 110031597A
- Authority
- CN
- China
- Prior art keywords
- experiment
- water quality
- fish
- water
- monitoring method
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/18—Water
- G01N33/186—Water using one or more living organisms, e.g. a fish
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Engineering & Computer Science (AREA)
- Food Science & Technology (AREA)
- Medicinal Chemistry (AREA)
- Physics & Mathematics (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of biological water monitoring method, is compensated using KCF track algorithm to the fish body moving target for colliding or blocking is generated in loss and motion process;Smooth trajectory is carried out using optical flow algorithm during tracking, fish locomotion track has been directly obtained, has avoided the center of mass point from image sequence and extract again, largely improve efficiency;What is extracted is three-dimensional track coordinate, provides more true fish swimming track, can reduce potential error of the fish towards video camera transverse movement as far as possible, convert visualization motion profile for 3D data, provide the visual representation method of fish locomotion;Traditional water quality method for monitoring abnormality based on SVM and XGBoost is merged with the monitoring method of deep learning neural network using the mode of integrated study, the error rate of water quality judging result exponentially declines, and accuracy rate is high.The present invention effectively saves human and material resources and financial resources, while enhancing the real-time and robustness of water quality monitoring.
Description
Technical field
The present invention relates to water quality monitoring field more particularly to it is a kind of utilize fish locomotion track carry out biological water monitoring
Method.
Background technique
Water is valuable source for the survival of mankind, and current water resources problems have become global problem.The mankind can develop
The water resource utilized is sharply being reduced, and with the continuous development of industry and city-building, industry and sanitary sewage are continuously increased, water
Matter pollution monitoring has a very important significance in water environment safety protection field.
For traditional water quality monitoring method mainly based on physico-chemical method, the relevant technologies are more mature and can be to water quality situation
Accurately judged.But physico-chemical method carries out physics and chemical analysis to it again by artificial acquisition water sample, and the process is time-consuming
Effort is unable to satisfy requirement of real-time.For this purpose, the water quality monitoring method based on biological monitoring be suggested to expectation realize have at
This is low, accuracy is high and has the water quality monitoring system of on-line monitoring with warning function.Biological monitoring mainly utilizes biology to ring
It is reacted caused by border pollution or environmental change directly or indirectly to embody the pollution condition of water quality.
Detection to piscine organism is using earliest one of Bio-monitoring method.At present using fish as the water of model organism
Quality supervision examining system be mainly pass through real time monitoring and analyze piscine organism individual or group behavior, by its kinematic parameter come into
Row judges the pollution level of water quality.It is existing research both in two-dimensional surface fish track move, practical Mesichthyes be
It moves about in three-dimensional space, is only the actual motion state that two-dimensional characteristic parameter can't accurately describe fish, the reality of water quality
Border situation cannot be effectively reflected to affect the management and improvement to water environment by monitoring result.
Summary of the invention
In view of the above technical problems, the present invention provides a kind of biological water monitoring methods, to solve above-mentioned background
Itd is proposed in technology can not long term monitoring, the one-sidedness problem of two dimensional motion, while saving human and material resources and financial resources, enhance water
The real-time and robustness that quality supervision is surveyed.
To achieve the above object, present invention employs following technical solutions: a kind of biological water monitoring method, including with
Lower step:
S1. Image Acquisition and pretreatment stage, comprising:
S11. fish that are sensitive to pollutant and being easy to raise are chosen as tested fingerling;
S12. control experiment is set up, experiment reagent used is added:
Blank assay and control experiment are set, the tested fingerling of blank assay and control experiment is put into experiment cylinder, to blank
The experimental water for not adding any pollutant is added in the experiment cylinder of experiment, experiment used is added into the experiment cylinder of control experiment
Reagent;
S13. Image Acquisition:
Movement of the tested fingerling in experiment cylinder is shot using two CCD cameras, a video camera is placed on experiment cylinder
Front, a video camera be placed on experiment cylinder surface;
S14. image preprocessing:
Collected video image is subjected to the corresponding points that Feature Points Matching finds different moments first with Hungary Algorithm, into
And tracked using Kalman filtering, KCF target tracking algorism carries out tracking target compensation, reaches after stablizing, and utilizes light stream
Algorithm progress motion profile is smooth, obtains two-dimentional track pixel coordinate;
S2. three-dimensional coordinate track synthesis phase, comprising:
S21. two-dimensional pixel coordinate is pre-processed, chooses continuous two-dimensional pixel coordinate points;
S22. it is distorted according to direct linear transformation's method (DLT) calibration camera;
S23. image corresponding points are determined;
S24. fish locomotion track voxel coordinate is calculated using corresponding points as match point;
S3. Classification Neural building, training stage, comprising:
S31. characteristic parameter extraction:
Speed, acceleration, curvature, five groups of center travelling distance, dispersion characteristic parameters are extracted from 3 D motion trace;
S32. characteristic parameter Database:
Integration is carried out to all characteristic parameter samples and establishes characteristic parameter database;
S33. traditional base Multiple Classifier Fusion:
It is based on the water quality exception monitoring model of support vector machines (SVM) and XGBoost using characteristic parameter Database, obtains
It can identify the base classifier of different quality;
S34. deep neural network model is established:
For space motion path pixel point set, pointnet network model is constructed, the space encoding for learning each point is then poly-
All features are closed to global characteristics;
S35. traditional base classifier and deep neural network model are integrated:
Using traditional water quality method for monitoring abnormality based on support vector machines (SVM) and XGBoost, with deep learning nerve net
The monitoring method of network is merged, i.e., completes learning tasks by building and in conjunction with multiple learners;
S36. model training obtains Water Quality Assessment Model:
Normal water quality and exception water quality sample set are established, sample set is sent into model as input and is trained, water quality is obtained
Evaluation model;
S4. the online water quality monitoring stage
The video sequence acquired in real time is judged using trained Water Quality Assessment Model, if there is exception water quality, is sent out
It alarms out, related personnel takes timely measure.
Red crucian and zebra fish are chosen in step S11 as tested fingerling.
It sets up three groups of experiments in step S12, is put into phase in the blank assay of every group of experiment and the experiment cylinder of control experiment
With the tested fingerling of quantity, the experimental water for not adding any pollutant is added in the experiment cylinder of the blank assay of every group of experiment,
The metrifonate aqueous solution that same volume solubility is 1.5%, second group of experiment is added in the experiment cylinder of the control experiment of first group of experiment
Control experiment experiment cylinder in be added same volume pH value be 3.5 oxalic acid solution, third group experiment control experiment reality
It tests and same volume is added in cylinderConcentration is 5.75 mg/L'sSolution.Further, experimental water is sufficiently to expose
The tap water of gas, 25 ± 1 DEG C of temperature, 8 ㎎ of dissolved oxygen >/L, illumination ratio 16h:8h(daytime: night), experiment cylinder is transparent
Fishbowl.
Shoal of fish video is acquired after experiment starts five minutes in step S13.
In step S35 using three classifiers (pointnet neural network model, SVM, XGBoost) in test sample
Classify, finally carries out final prediction using ballot method and export.
Model training the number of iterations is 100 times in step S36, and stochastic gradient descent is used in optimization algorithm, was being trained
Different learning rates is set in journey, and the number of iterations is that learning rate is 1e-3 between 0-65, and learning rate is 1e-4,86- between 66-85
Learning rate is 1e-5 between 99.
In step S4 during water quality monitoring, using the distribution proportion of feedback self correcting system balance sample classification, no
Disconnected corrective networks, provide a kind of training method of sustainable study.
Judging result not clock synchronization, is modified update to model in step S4.
Compared with prior art, the invention has the benefit that using KCF track algorithm to being produced in loss and motion process
Raw collision or the fish body moving target blocked compensate, and significantly reduce illumination, block, brought by electrical device noise
Error;Smooth trajectory is carried out using optical flow algorithm during tracking, fish locomotion track has been directly obtained, has avoided from image
Center of mass point is extracted again in sequence, largely improves efficiency;What is extracted is three-dimensional track coordinate, is provided more true
Fish swimming track, potential error of the fish towards video camera transverse movement can be reduced as far as possible, convert 3D data to
Motion profile is visualized, the visual representation method of fish locomotion is provided;Using the mode of integrated study by traditional based on branch
The water quality method for monitoring abnormality for holding vector machine (SVM) and XGBoost is merged with the monitoring method of deep learning neural network,
The error rate of water quality judging result exponentially declines, and accuracy rate is high.The present invention effectively saves human and material resources and financial resources, together
When enhance the real-time and robustness of water quality monitoring.
Detailed description of the invention
Fig. 1 is a kind of overall flow figure of biological water monitoring method of the invention;
Fig. 2 is three-dimensional track pixel coordinate synthetic schemes of the invention.
Specific embodiment
To make to have further understanding to the purpose of the present invention, construction, feature and its function, hereby cooperate embodiment detailed
It is described as follows.
Please refer to Fig. 1, a kind of biological water monitoring method of the invention, including Image Acquisition and pretreatment stage, three
Tie up Grid Track synthesis phase, Classification Neural building, training stage and online water quality monitoring stage.
1. Image Acquisition and pretreatment stage, comprising the following steps:
Step 1. chooses fish that are sensitive to pollutant and being easy to raise as tested fingerling
The key factor for choosing fish progress water quality monitoring can be divided into the following: 1, when environmental change and pollutant are invaded
The sensibility and alertness of fish;2, the shoal of fish should be distributed greatly and widely and be easy to capture;3, fish are easy to feed in the lab
It supports.After consulting pertinent literature and fish breeding is considered the problems of, based on red crucian and zebra fish its generality, easily raising, numerous
Grow ability it is strong, it is cheap, low advantage is required to production environment, and zebra fish is widely used in grinding in field of genetic engineering
Study carefully vertebrate evolution and human inheritance's disease, is a kind of typical fish, under lab, zebra fish is often used as toxicity
The research of test, therefore select red crucian and zebra fish as tested fingerling;
Step 2. sets up control experiment, and experiment reagent used is added
Experiment is divided into three groups, and using the method for parallel laboratory test, only water quality pollutant is different for three groups of experiments, and other conditions are kept completely
Unanimously.Experiment porch is 50 ㎝ of diameter, the round of high 40 ㎝ tests cylinder.Every group of setting blank assay and multiple control experiments, will
Tested fingerling is put into blank assay and the experiment cylinder of control experiment.It is added in the experiment cylinder of first group of blank assay certain
Volume does not add the experimental water of any pollutant, is separately added into same volume various concentration in the experiment cylinder of multiple control experiments
Metrifonate aqueous solution;The reality that certain volume does not add any pollutant is added in the experiment cylinder of second group of blank assay
It tests and uses water, the oxalic acid solution of same volume different PH is separately added into the experiment cylinder of multiple control experiments;In the sky of third group
Certain volume is added in the experiment cylinder of white experiment and does not add the test water of any pollutant, in the experiment cylinder of multiple control experiments
It is separately added into same volume differenceConcentrationSolution.
Step 3. Image Acquisition
For continuous automatic, objectively and accurately record and analysis fish various actions, as shown in Fig. 2, experiment uses two
CCD camera tracks the fish in fish jar, is detected, and in the transmission of video images to primary server taken, passes through main clothes
Device be engaged in fish progress behavioural analysis.Experiment experiment cylinder used is transparent fishbowl, and two CCD cameras are respectively in fish
The front and surface of cylinder, they obtain fish from two different visual angles simultaneously and run video image.It is adopted in experimentation
Fish body motion video image is obtained with CCD camera, carries out shoal of fish video image acquisition, every frame image after experiment starts five minutes
Size is 480 × 640, and frame speed is 25 frames/second.
Due to having differences between fish individual, the motor behavior of every fish can change with the variation of time and environment,
In order to guarantee objectivity, the accuracy of experiment, experimental data is acquired according to following principle :(a) experimental water is sufficiently aerated
Tap water, 25 ± 1 DEG C of temperature, 8 ㎎ of dissolved oxygen >/L, illumination ratio 16h:8h(daytime: night), the water tested in cylinder guarantees one
It 24 hours flow, flow velocity be subject to do not influence fish normal behaviour activity;(b) time for acquiring data is in continuous one day
16 hours daytimes;(c) regular time gives fish feeding daily.To the red crucian of the different age of a fish and zebra fish according to step 2
Mode is tested, and makes fish is (identical as blank assay environment) in experiment cylinder and experimental water environment to adapt to before drug exposure
It 3 days, adapts to environment or container to eliminate fish and changes bring behavior change.
Step 4: image preprocessing
The complex environment of dynamic and multiple target in water, when two-dimentional track pixel coordinate is reconstructed into three-dimensional track pixel coordinate, water
Error that optical index is different from the optical index of air and occurs etc. is caused to using computer vision progress water quality monitoring
Greatly interference.Feature Points Matching is carried out using Hungary Algorithm, finds the corresponding points of different moments, and then filter with Kalman
Wave tracks the coordinate and size of target, then carries out smooth trajectory using optical flow algorithm.Due to tracking during unstability,
It is compensated using KCF track algorithm to the fish body moving target for colliding or blocking is generated in loss and motion process.
2. three-dimensional coordinate track synthesis phase
Three-dimensional track moves compared with traditional two-dimentional trajectory time sequence, and three-dimensional track movement can provide more true fish
Motion profile can reduce potential error of the fish towards video camera transverse movement as far as possible, convert visualization for 3D data
Motion profile provides the visual representation method of fish locomotion for experimental study.
If Stereo matching and tracking handled as two mutually independent processes, when the number of fish school increases
When, every fish is close to each other during travelling or the frequency blocked is very high, causes the two-dimentional path segment extracted universal
Shorter, the uncertainty of Stereo matching is difficult to be eliminated using kinetic characteristic in a short time.If the center of mass point for the fish extracted
It is superimposed upon overlong time close when one piece or travelling, the track serious loss of center of mass point synthesis is will lead to, in turn results in 3D
The error of reconstruction influences the effect rebuild, it can be seen that reconstruction process and tracking process are complementary.In the present invention
In, as a whole with real-time tracking by three-dimensional point building.In reconstruction process, using two video cameras in different angle pair
Fish are shot, if center of mass point a visual angle it is close or superposition, can jump to another visual angle eliminate track following
Uncertainty, such each time point can extract fish body center-of-mass coordinate, largely by the difficulty of characteristic matching
It is reduced.
Referring to figure 2., experimental provision belongs to monocular multi-view imaging, and when design determines the several of two video cameras and fish jar
What relationship can obtain the center of mass point position coordinates consistent from two view fishes according to these geometrical relationships fixed.
According to fig. 2 shown in experimental rig, the image pixel coordinates that front video camera takes are the image pixel that surface takes
Coordinate is, wherein described is the pixel coordinate in same direction, so the information of two camera shooting arbors are overlappings.It is logical
Cross two CCD cameras acquisition multiframe consecutive images, two cameras must assure that be it is synchronous, mistake could be reduced as far as possible in this way
It is influenced brought by poor.It is distorted according to direct linear transformation's method (DLT) calibration camera, image corresponding points is determined, by corresponding points
The voxel coordinate of fish locomotion track is calculated as match point.
3. Classification Neural building, training stage, comprising the following steps:
Step 1. characteristic parameter extraction
Under normal water quality, speed, acceleration, curvature presentation change by a small margin, and average travelling does not occur apart from roughly the same
The situation of big ups and downs, dispersion is also relatively small, and situation is stablized, and each Parameters variation is larger under exception water quality.Cause
This extracts speed, acceleration, curvature, five groups of center travelling distance, dispersion characteristic parameters from 3 D motion trace.
Step 2. characteristic parameter Database
Integration is carried out to all characteristic parameter samples and establishes characteristic parameter database.
Step 3. tradition base Multiple Classifier Fusion
Using in SVM and XGBoost assorting process, the Fish behavior kinematic feature factor for being able to reflect change of water quality is calculated
Data carry out certain processing to data, keep kinematic parameter feature more prominent;It is based on using characteristic parameter Database
The water quality exception monitoring model of SVM and XGBoost;Finally the Fish behavior characteristic parameter under unknown water quality is carried out using model
Assay achievees the purpose that water quality abnormality detection.
Step 4. deep neural network model is established
On the basis of extracting fish locomotion behavioural characteristic and calculating feature, its quantitative relationship with water quality safety is probed into, from
Essentially the fish under normal water quality and exception water quality have the motion feature of significant difference, are considered as the different fortune of two classes
Dynamic model formula, therefore water quality security evaluation is a classification problem.Classification problem is that target is classified with non-targeted, wherein non-
Target refers to all objects (non-targeted in the present invention is Fish behavior parameter in different pollution level classifications) for removing target,
It can not be described with one or several modes.
For space motion path pixel point set, pointnet network model is constructed, the space encoding for learning each point is right
After polymerize all features to global characteristics.Local feature is extracted from small field using disaggregated model, is then further processed production
Raw higher level feature, this process are repeated continuously the global characteristics until obtaining entire point set.
Step 5. tradition base classifier and deep neural network model are integrated
Fish locomotion trace image has different feature extracting methods, and every kind of method all stresses the description of image different aspect, will
These features combine, can expression characteristic information more fully hereinafter.Using it is traditional based on support vector machines (SVM) and
The water quality method for monitoring abnormality of XGBoost, is merged with the monitoring method of deep learning neural network, i.e., by constructing and tying
Multiple learners are closed to complete learning tasks.In Integrated Strategy, the monitoring of water quality normal anomaly can regard two classification problems as, use three
Classifier (pointnet neural network model, SVM, XGBoost) is classified in test sample, and wherein √ presentation class is being just
Really, × presentation class mistake.It finally carries out final prediction using ballot method to export, i.e., " the minority is subordinate to the majority ".
Step 6. model training
Normal water quality and exception water quality sample set are established, sample set is sent into model as input and is trained, water quality is obtained
Evaluation model.In an experiment using the experimental water and addition metrifonate, oxalic acid solution, potassium bichromate for being added without any pollutant
Three groups of fish locomotion track datas being obtained of experimental water be trained.The number of iterations is set as 100, with the number of iterations
Increase, training accuracy rate be also stepped up.Stochastic gradient descent (SGD) is used in optimization algorithm, a pass in SGD algorithm
Bond parameter is learning rate, and different learning rates is arranged in the training process, is equal to 1e-3,66- between 0-65 in the number of iterations
It is equal to 1e-4 between 85, is equal to 1e-5 between 86-99.Accuracy rate highest can achieve 95% or so, that is, the pointnet mould selected
Type complies fully with water quality assessment requirement.
4. the online water quality monitoring stage
The video sequence acquired in real time is judged using trained Water Quality Assessment Model, if there is exception water quality in judgement,
Alarm is then issued, related personnel takes timely measure.(for example water quality is used when related personnel detects water quality actual conditions
Detector) Water Quality Assessment Model judging result is incorrect when there is false alarm for discovery afterwards, update is modified to model.
Water quality monitoring is a long-term observation behavior, for fixed water environment, provided that giving machine learning
Set of data samples be unbalanced class sample, i.e., occur the sample size of normal water quality and the sample of exception water quality in monitoring process
This quantitative proportion gap great disparity, then it is unavailable that the model trained deviation will occur.During water quality monitoring, using anti-
The distribution proportion of self correcting system balance sample classification is presented, continuous corrective networks provide a kind of training method of sustainable study,
Accuracy rate is improved to reduce error with this, achievees the purpose that real-time monitoring.
A kind of biological water monitoring method of the invention, using KCF track algorithm to being generated in loss and motion process
The fish body moving target for colliding or blocking compensates, and significantly reduces illumination, blocks, misses brought by electrical device noise
Difference;Smooth trajectory is carried out using optical flow algorithm during tracking, fish locomotion track has been directly obtained, has avoided from image sequence
Center of mass point is extracted again in column, largely improves efficiency;What is extracted is three-dimensional track coordinate, is provided more true
Fish swimming track can reduce potential error of the fish towards video camera transverse movement as far as possible, and converting 3D data to can
Depending on changing motion profile, the visual representation method of fish locomotion is provided;Using the mode of integrated study by traditional based on support
The water quality method for monitoring abnormality of vector machine (SVM) and XGBoost are merged with the monitoring method of deep learning neural network, water
The error rate of matter judging result exponentially declines, and accuracy rate is high.The present invention effectively saves human and material resources and financial resources, simultaneously
Enhance the real-time and robustness of water quality monitoring.
The present invention is described by above-mentioned related embodiment, however above-described embodiment is only to implement example of the invention.
It must be noted that the embodiment disclosed is not limiting as the scope of the present invention.On the contrary, do not depart from spirit of the invention and
It is changed and retouched made by range, belongs to scope of patent protection of the invention.
Claims (9)
1. a kind of biological water monitoring method, which comprises the following steps:
S1. Image Acquisition and pretreatment stage, comprising:
S11. fish that are sensitive to pollutant and being easy to raise are chosen as tested fingerling;
S12. control experiment is set up, experiment reagent used is added:
Blank assay and control experiment are set, the tested fingerling of blank assay and control experiment is put into experiment cylinder, to blank
The experimental water for not adding any pollutant is added in the experiment cylinder of experiment, is added used in research into the experiment cylinder of control experiment
Experiment reagent;
S13. Image Acquisition:
Movement of the tested fingerling in experiment cylinder is shot using two CCD cameras, a video camera is placed on experiment cylinder
Front, a video camera be placed on experiment cylinder surface;
S14. image preprocessing:
Collected video image is subjected to the corresponding points that Feature Points Matching finds different moments first with Hungary Algorithm, into
And tracked using Kalman filtering, KCF target tracking algorism carries out tracking target compensation, reaches after stablizing, and utilizes light stream
Algorithm progress motion profile is smooth, obtains two-dimentional track pixel coordinate;
S2. three-dimensional coordinate track synthesis phase, comprising:
S21. two-dimensional pixel coordinate is pre-processed, chooses continuous two-dimensional pixel coordinate points;
S22. it is distorted according to direct linear transformation's method calibration camera;
S23. image corresponding points are determined;
S24. fish locomotion track voxel coordinate is calculated using corresponding points as match point;
S3. Classification Neural building, training stage, comprising:
S31. characteristic parameter extraction:
Speed, acceleration, curvature, five groups of center travelling distance, dispersion characteristic parameters are extracted from 3 D motion trace;
S32. characteristic parameter Database:
Integration is carried out to all characteristic parameter samples and establishes characteristic parameter database;
S33. traditional base Multiple Classifier Fusion:
Water quality exception monitoring model using characteristic parameter Database based on SVM and XGBoost, obtains to identify difference
The base classifier of water quality;
S34. deep neural network model is established:
For space motion path pixel point set, pointnet network model is constructed, the space encoding for learning each point is then poly-
All features are closed to global characteristics;
S35. traditional base classifier and deep neural network model are integrated:
Monitoring method using traditional water quality method for monitoring abnormality based on SVM and XGBoost, with deep learning neural network
It is merged, i.e., completes learning tasks by building and in conjunction with multiple learners;
S36. model training obtains Water Quality Assessment Model:
Normal water quality and exception water quality sample set are established, sample set is sent into model as input and is trained, water quality is obtained
Evaluation model;
S4. online water quality monitoring stage:
The video sequence acquired in real time is judged using trained Water Quality Assessment Model, if there is exception water quality, is sent out
It alarms out, related personnel takes timely measure.
2. a kind of biological water monitoring method as described in claim 1, which is characterized in that choose red crucian in step S11
With zebra fish as tested fingerling.
3. a kind of biological water monitoring method as described in claim 1, which is characterized in that set up three groups of realities in step S12
It tests, the tested fingerling of identical quantity is put into the blank assay of every group of experiment and the experiment cylinder of multiple control experiments, every group real
The experimental water for not adding any pollutant, multiple control experiments of first group of experiment are added in the experiment cylinder for the blank assay tested
Experiment cylinder in be separately added into the metrifonate aqueous solution of same volume various concentration, the reality of multiple control experiments of second group of experiment
It tests in cylinder and is separately added into the oxalic acid solution of same volume different PH, divide in the experiment cylinder of multiple control experiments of third group experiment
It Jia Ru not same volume differenceConcentrationSolution.
4. a kind of biological water monitoring method as claimed in claim 3, which is characterized in that the experimental water is sufficiently to expose
The tap water of gas, 25 ± 1 DEG C of temperature, 8 ㎎ of dissolved oxygen >/L, illumination ratio 16h:8h, the experiment cylinder is transparent glass fish
Cylinder.
5. a kind of biological water monitoring method as described in claim 1, which is characterized in that start in step S13 in experiment
Shoal of fish video is acquired after five minutes.
6. a kind of biological water monitoring method as described in claim 1, which is characterized in that used in step S35
These three classifiers of pointnet neural network model, SVM and XGBoost are classified in test sample, finally using throwing
Ticket method carries out final prediction output.
7. a kind of biological water monitoring method as described in claim 1, which is characterized in that model training changes in step S36
Generation number is 100 times, and stochastic gradient descent is used in optimization algorithm, and different learning rates, iteration time are arranged in the training process
Number learning rate between 0-65 is 1e-3, and learning rate is 1e-4 between 66-85, and learning rate is 1e-5 between 86-99.
8. a kind of biological water monitoring method as described in claim 1, which is characterized in that in water quality monitoring in step S4
Cheng Zhong, using the distribution proportion of feedback self correcting system balance sample classification, continuous corrective networks provide a kind of sustainable study
Training method.
9. a kind of biological water monitoring method as described in claim 1, which is characterized in that judging result is not right in step S4
When, update is modified to Water Quality Assessment Model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910316179.4A CN110031597A (en) | 2019-04-19 | 2019-04-19 | A kind of biological water monitoring method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910316179.4A CN110031597A (en) | 2019-04-19 | 2019-04-19 | A kind of biological water monitoring method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110031597A true CN110031597A (en) | 2019-07-19 |
Family
ID=67239219
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910316179.4A Pending CN110031597A (en) | 2019-04-19 | 2019-04-19 | A kind of biological water monitoring method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110031597A (en) |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110633530A (en) * | 2019-09-18 | 2019-12-31 | 南通大学 | Fishway design method based on computational fluid dynamics and convolutional neural network |
CN110645990A (en) * | 2019-10-17 | 2020-01-03 | 浙江科技学院 | Water area cruising system based on SVM and Kalman filtering for fish school dynamic prediction |
CN111476765A (en) * | 2020-03-30 | 2020-07-31 | 深圳市水务(集团)有限公司 | Water quality judging method and device |
CN112070799A (en) * | 2020-05-29 | 2020-12-11 | 清华大学 | Fish trajectory tracking method and system based on artificial neural network |
CN112946226A (en) * | 2021-04-02 | 2021-06-11 | 墣锦环境工程(海南)有限公司 | Three-dimensional visual simulation method for soil and underground water pollution |
CN113063913A (en) * | 2021-03-29 | 2021-07-02 | 广东骏信科技有限公司 | Water toxicity biological monitor and monitoring method |
CN113111480A (en) * | 2021-02-22 | 2021-07-13 | 同济大学 | Method and device for diagnosing and detecting running state of drainage pipe network |
CN113516635A (en) * | 2021-06-15 | 2021-10-19 | 中国农业大学 | Fish-vegetable symbiotic system and vegetable nitrogen element demand estimation method based on fish behaviors |
CN113984763A (en) * | 2021-10-28 | 2022-01-28 | 内蒙古大学 | Visual identification-based insect repellent pesticide effect experimental device and method |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104749334A (en) * | 2015-02-16 | 2015-07-01 | 燕山大学 | Mode-recognition-based design method for biological abnormal water quality evaluation system |
CN108053427A (en) * | 2017-10-31 | 2018-05-18 | 深圳大学 | A kind of modified multi-object tracking method, system and device based on KCF and Kalman |
-
2019
- 2019-04-19 CN CN201910316179.4A patent/CN110031597A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104749334A (en) * | 2015-02-16 | 2015-07-01 | 燕山大学 | Mode-recognition-based design method for biological abnormal water quality evaluation system |
CN108053427A (en) * | 2017-10-31 | 2018-05-18 | 深圳大学 | A kind of modified multi-object tracking method, system and device based on KCF and Kalman |
Non-Patent Citations (4)
Title |
---|
SHUHONG CHENG: "Water Quality Anomaly Monitoring Based on Kalman Filter and Convolution Neural Network", 《2019 INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE AND COMPUTING SCIENCE》 * |
SHUHONG CHENG: "Water Quality Monitoring Method Based on TLD 3D Fish Tracking and XGBoost", 《MATHEMATICAL PROBLEMS IN ENGINEERING》 * |
钱婷 等: "三维激光扫描点云分类方法研究进展综述", 《电脑与信息技术》 * |
陈海宏: "《机器学习原理及应用》", 31 July 2017, 电子科技大学出版社 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110633530A (en) * | 2019-09-18 | 2019-12-31 | 南通大学 | Fishway design method based on computational fluid dynamics and convolutional neural network |
CN110633530B (en) * | 2019-09-18 | 2023-05-05 | 南通大学 | Fishway design method based on computational fluid dynamics and convolutional neural network |
CN110645990A (en) * | 2019-10-17 | 2020-01-03 | 浙江科技学院 | Water area cruising system based on SVM and Kalman filtering for fish school dynamic prediction |
CN111476765A (en) * | 2020-03-30 | 2020-07-31 | 深圳市水务(集团)有限公司 | Water quality judging method and device |
CN112070799A (en) * | 2020-05-29 | 2020-12-11 | 清华大学 | Fish trajectory tracking method and system based on artificial neural network |
CN113111480A (en) * | 2021-02-22 | 2021-07-13 | 同济大学 | Method and device for diagnosing and detecting running state of drainage pipe network |
CN113111480B (en) * | 2021-02-22 | 2022-07-29 | 同济大学 | Method and device for diagnosing and detecting running state of drainage pipe network |
CN113063913A (en) * | 2021-03-29 | 2021-07-02 | 广东骏信科技有限公司 | Water toxicity biological monitor and monitoring method |
CN112946226A (en) * | 2021-04-02 | 2021-06-11 | 墣锦环境工程(海南)有限公司 | Three-dimensional visual simulation method for soil and underground water pollution |
CN113516635A (en) * | 2021-06-15 | 2021-10-19 | 中国农业大学 | Fish-vegetable symbiotic system and vegetable nitrogen element demand estimation method based on fish behaviors |
CN113516635B (en) * | 2021-06-15 | 2024-02-27 | 中国农业大学 | Fish and vegetable symbiotic system and vegetable nitrogen element demand estimation method based on fish behaviors |
CN113984763A (en) * | 2021-10-28 | 2022-01-28 | 内蒙古大学 | Visual identification-based insect repellent pesticide effect experimental device and method |
CN113984763B (en) * | 2021-10-28 | 2024-03-26 | 内蒙古大学 | Insect repellent efficacy experimental device and method based on visual recognition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110031597A (en) | A kind of biological water monitoring method | |
Liu et al. | YOLOv5-Tassel: Detecting tassels in RGB UAV imagery with improved YOLOv5 based on transfer learning | |
Murugesan et al. | Efficient anomaly detection in surveillance videos based on multi layer perception recurrent neural network | |
Hamilton | Quantification and its applications in fluorescent microscopy imaging | |
CN105069746B (en) | Video real-time face replacement method and its system based on local affine invariant and color transfer technology | |
CN105354548B (en) | A kind of monitor video pedestrian recognition methods again based on ImageNet retrievals | |
Mohamed et al. | Msr-yolo: Method to enhance fish detection and tracking in fish farms | |
Nattkemper et al. | A neural classifier enabling high-throughput topological analysis of lymphocytes in tissue sections | |
CN108764308A (en) | A kind of recognition methods again of the pedestrian based on convolution loop network | |
CN109800689A (en) | A kind of method for tracking target based on space-time characteristic fusion study | |
CN102915446A (en) | Plant disease and pest detection method based on SVM (support vector machine) learning | |
CN109341580A (en) | Method for tracing, system and the device of profile of steel rail | |
CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
Zhan et al. | PDAM–STPNNet: A small target detection approach for wildland fire smoke through remote sensing images | |
CN103854027A (en) | Crowd behavior identification method | |
CN103344583A (en) | Praseodymium-neodymium (Pr/Nd) component content detection system and method based on machine vision | |
CN109934042A (en) | Adaptive video object behavior trajectory analysis method based on convolutional neural networks | |
CN110222832A (en) | Entrance of Changjiang River salt marshes macrobenthos habitat simulation prediction technique | |
CN105683355A (en) | Cell analysis system, cell analysis program, and cell analysis method | |
CN111898677A (en) | Plankton automatic detection method based on deep learning | |
Liu et al. | Automated classification of stems and leaves of potted plants based on point cloud data | |
CN112037252A (en) | Eagle eye vision-based target tracking method and system | |
Yasir et al. | ShipGeoNet: SAR image-based geometric feature extraction of ships using convolutional neural networks | |
CN107437091A (en) | Multilayer limits the positive and negative class change detecting method of SAR image of Boltzmann machine | |
Zhou et al. | Fish behavior analysis based on computer vision: a survey |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190719 |