CN104749334A - Mode-recognition-based design method for biological abnormal water quality evaluation system - Google Patents

Abstract

A method for designing a biological abnormal water quality evaluation system based on pattern recognition. The basic idea is to address the multi-factor, high-dimensional, nonlinear characteristics of water quality evaluation and the complex relationship between evaluation factors and water quality. , the present invention takes the common biological indicator fish in biological water quality evaluation as the research object, and uses computer image processing technology to propose a fast abnormal water quality evaluation system. First, based on computer vision, fish motion video images that can reflect water quality conditions are collected, and then image processing technology is used to obtain water quality evaluation indicators, that is, fish behavior characteristic parameters. At the same time, a semantic map that can reflect water quality conditions and fish behavior characteristic parameters is established based on pattern recognition. model, and finally examine the feasibility of the model. The invention can overcome the influence of environmental illumination changes and noise changes when extracting fish movement information, utilizes a support vector machine (SVM) to construct a water quality abnormal evaluation model and updates it in real time to improve model robustness.

Description

Design method of biological abnormal water quality evaluation system based on pattern recognition

technical field

The present invention relates to the field of computer vision and pattern recognition, especially a biological water quality system that successfully overcomes the shortcomings of traditional water quality evaluation such as multi-factor, high-dimensional, non-linear and the complex relationship between evaluation factors and water quality. evaluation method.

technical background

Water is the source of life, an important natural resource for our survival, and an important material basis for today's social and economic development. However, with the rapid development of urbanization and industrialization, countries all over the world are basically facing the challenges of water shortage and serious pollution. Therefore, water quality evaluation plays an important role in water resources management, protection and planning, but how to evaluate water quality accurately, objectively and scientifically is still a difficult problem. At present, most scholars are based on the comprehensive evaluation of water quality pollution factors (evaluation indicators) collected by physical and chemical analysis technology and automatic detection technology to detect and analyze water quality, and then obtain the water quality status. However, when testing water quality in this way, first of all, the evaluation data has shortcomings such as multiple factors, nonlinearity, and high dimensionality, so that it is difficult to establish a model for quickly identifying water quality; secondly, the evaluation method is either highly subjective or cannot be comprehensive. Deeply reflect the water quality status. In recent years, biological detection technology has attracted people's attention in water quality evaluation due to its advantages of comprehensiveness, enrichment, continuity, high response sensitivity, and intuitiveness. In biological water quality assessment, fishes are regarded as important indicator organisms due to their advantages of easy identification and obvious movement characteristics. Fish and the aquatic ecological environment in which they live constitute an interactive unified system, and the two are interdependent and co-evolved. The unity and co-evolution of the water environment and fish are the basis for water quality evaluation, and the various ecological responses of fish to changes in the water environment are the basis for water quality evaluation. Therefore, the use of fish behavior characteristics as evaluation indicators can not only involve fewer indicators but also reflect the potential impact of mixed water pollution.

Computer vision mainly uses computers to simulate human visual functions, extracts information from images of objective things, processes and understands them, and finally uses them in actual detection, measurement and control. Therefore, many scholars at home and abroad have successfully extracted fish motion characteristic parameters based on computer vision technology. However, only a few scholars have established water quality evaluation models based on these parameters, interpreting the qualitative relationship between the aquatic ecological environment and fish. Cigdem et al. proposed a rule-based trajectory filtering mechanism to filter the trajectory monitored by computer vision technology, and successfully extracted the abnormal trajectory of fish. This method is expected to be applied to the understanding of the relationship between fish behavior and water quality. Especially in the field of normal and abnormal water quality detection. By defining the state of fish movement behavior, Carlos et al. used the background difference technology to obtain the different points of fish in polluted and unpolluted water, and then used the recursive graph algorithm to draw the trajectory, defined the descriptor based on the recursive graph, and obtained a time-lapse According to the binary recursive vector matrix, the abnormal water quality added with chlorpyrifos pesticide was successfully judged according to the different points of the matrix. Liao et al. used support vector machine (SVM) to learn and test the behavior of zebrafish under different copper ion concentrations, and studied the test accuracy under different kernel function types. The research showed that this method can effectively evaluate abnormal water quality. Although both Carlos and Liao evaluated the water quality status by extracting parameters related to fish trajectories, they did not give a specific water quality evaluation system process. Cheng Shuhong and others established a water quality anomaly detection model based on computer vision and SVM. The experimental results show that the model can quickly and effectively detect water quality anomalies, but the robustness of the model has not been verified, especially the anti-interference of light or external irrelevant events. Interfering, and only a scheme for building a model is given, and a complete water quality anomaly evaluation system is also not given.

In summary, biological water quality evaluation plays a vital role in the field of water quality safety testing, providing a theoretical basis for accurate and timely detection of water quality anomalies. However, how to design a perfect abnormal water quality evaluation system and investigate the robustness of the system is the key point of the present invention.

Contents of the invention

The purpose of the present invention is (1) to design a complete abnormal water quality evaluation system; (2) to improve the robustness of the system, especially for the impact of ambient light changes and noise changes; (3) to verify the feasibility of the system.

The basic idea of the present invention is: Aiming at the multi-factor, high-dimensional, nonlinear and other characteristics of water quality evaluation, and the complex relationship between evaluation factors and water quality, the present invention uses biological indicators commonly used in biological water quality evaluation Taking animal fish as the research object, a rapid abnormal water quality evaluation system is proposed by using computer image processing technology. First, based on computer vision, fish motion video images that can reflect water quality conditions are collected, and then image processing technology is used to obtain water quality evaluation indicators, that is, fish behavior characteristic parameters. At the same time, a semantic map that can reflect water quality conditions and fish behavior characteristic parameters is established based on pattern recognition. model, and finally examine the feasibility of the model.

In order to solve the above-mentioned technical problems and realize the above-mentioned purpose, the present invention is achieved through the following technical solutions:

A method for designing a biological abnormal water quality evaluation system based on pattern recognition, which includes the following steps:

(1) Build a water quality evaluation platform

Build a water quality evaluation platform, and use computer vision technology to convert live fish movements into video image data;

(2) Image preprocessing

Image processing technology is used to preprocess the acquired image. First, bilateral filtering is performed on the video frame image to remove noise while retaining edge information, and then perform DCT transformation on the filtered image to eliminate the influence of illumination;

(3) Detection and tracking of fish moving targets

Since the Kalman filter has the advantages of less calculation and real-time tracking of the tracking algorithm, it is used to track the fish and obtain the trajectory of the fish;

(4) Extraction of water quality evaluation indicators

Obtain the water quality evaluation index by tracking the movement trajectory of the obtained fish, namely the four parameters of swimming speed, swimming distance, swimming acceleration and direction of rotation angle;

(5) Establish a characteristic parameter database

Preprocess the extracted motion parameters under normal and abnormal water quality, remove gross errors, obtain normalized data, and establish a characteristic parameter database;

(6) Establishment of water quality evaluation model

In order to quickly establish a water quality evaluation model, firstly, PCA dimensionality reduction is performed on the acquired evaluation indicators, and then the support vector machine (SVM) is used to construct a water quality anomaly evaluation model;

(7) Investigate the feasibility of the water quality evaluation system

In order to verify the feasibility of the design method of the biological abnormal water quality evaluation system based on pattern recognition, an environmental detector is used to verify whether the evaluation result is correct when evaluating the abnormal water quality;

(8) Online water quality evaluation

Design an online abnormal water quality evaluation system, extract fish movement behavior parameters in water quality online, and use the water quality evaluation model to evaluate water quality anomalies; at the same time, use the evaluation results to update the characteristic parameter database and evaluation model to improve the robustness of the model.

Due to the adoption of the above-mentioned technical solution, the design method of a biological abnormal water quality evaluation system based on pattern recognition provided by the present invention has such beneficial effects compared with the prior art:

The invention designs a complete abnormal water quality evaluation system and improves the robustness of the system, especially for the influence of ambient light changes and noise changes, and has a certain theoretical basis and reference value for online water quality evaluation.

The invention can overcome the influence of environmental light changes and noise changes when extracting fish movement information, and uses Kalman filter to accurately track fish movement trajectories to extract movement parameters to construct feature matrix, and uses support vector machine (SVM) to construct water quality anomaly evaluation model and real-time Updates improve model robustness.

Description of drawings

Fig. 1 is a flow chart of the design method of the biological abnormal water quality evaluation system based on pattern recognition;

Figure 2 is a schematic diagram of the water quality evaluation platform built;

Fig. 3 is the flowchart of image preprocessing;

Fig. 4 is the result figure after image preprocessing;

Fig. 5 is the trajectory diagram of real-time tracking of fish based on Kalman filter;

Fig. 6 is a flow chart of the water quality evaluation model;

Fig. 7 is a schematic diagram of an online abnormal water quality evaluation system.

Detailed ways

Below in conjunction with accompanying drawing and specific embodiment the present invention is described in further detail:

A flow chart of the design method of a biological abnormal water quality evaluation system based on pattern recognition proposed by the present invention is shown in Figure 1, and the following is divided into several parts to introduce specific implementation methods:

(1) Build a water quality evaluation platform

A water quality detection platform was built, and computer vision technology was used to convert the movement of live fish into video image data. The result of the construction is shown in Figure 2.

(2) Image preprocessing

The flow chart of image preprocessing is shown in Figure 3, where the original RGB image is preprocessed—after grayscale processing is shown in Figure 4(a), and the result of noise elimination after bilateral filtering is shown in Figure 4(b) As shown, the result after DCT domain enhancement is shown in Fig. 4(c).

As can be seen from Fig. 4, the image preprocessing method bilateral filtering proposed by the present invention can eliminate noise and the edges are also relatively clear, and the DCT transformation can effectively suppress the influence of the lighting environment. The preprocessing in this step can improve the robustness of the water quality evaluation system, especially for the impact of ambient light changes and noise changes.

(3) Detection and tracking of fish moving targets

Using the image preprocessed in step (2), the fish is tracked in real time based on the Kalman filter, and the trajectory of the fish is obtained. The tracking result is shown in Figure 5.

(4) Extraction of water quality evaluation indicators

Use step 3 to track the movement trajectories of fish under normal water quality and abnormal water quality, and extract the movement parameters that can reflect the water quality conditions: swimming speed, swimming distance, swimming acceleration, and direction of rotation angle.

(5) Establish a characteristic parameter database

The motion parameters under normal and abnormal water quality extracted in step 4 are preprocessed to eliminate gross errors, obtain normalized data, and establish normal and abnormal characteristic parameter databases respectively.

(6) Establishment of water quality evaluation model

In order to quickly establish a water quality evaluation model, firstly, PCA dimensionality reduction is performed on the obtained evaluation indicators, and some data are selected from the characteristic parameter database established in step (5) as the training sample set of the support vector machine (SVM), and the obtained evaluation indicators Carry out PCA dimension reduction processing, and then use support vector machine (SVM) to construct an evaluation model of water quality anomaly. As a training sample set, in order to quickly establish a water quality evaluation model and reduce data redundancy, PCA is used to reduce the dimensionality of the selected sample data, and the data after dimensionality reduction is used as the input of SVM, and the parameters of SVM are optimized. The water quality anomaly evaluation model is obtained.

Table 1 The recognition rate of abnormal water quality assessment of different models based on SVM (%)

Table 1 is the water quality evaluation model based on SVM and the accuracy rate of abnormal water quality test results, in which models 1, 2, 3, and 4 are models established by using the training sample sets of single fish 1, 2, 3, and 4, and model 5 is A water quality evaluation model based on the combination of training sample sets in single fish1,2,3,4. From the data on the diagonal line of Table 1 (marked in bold in the figure), it can be seen that the model established with the data of a single fish itself is significantly better than the results of the test set of the other three fish. As a result, there are differences between individual fish and the model established by a single fish is not robust. Although the test results of the test sample sets 1, 2, 3, and 4 of a single fish using model 5 are not as good as the test results of the model established by the individual fish's own data, it is better than the model established by a single fish for the other 3 fish The test results of the test set and the test results are all higher than 84%, so the model established by using the data of multiple fish has better robustness.

(7) Investigate the feasibility of the water quality evaluation system

In order to test the feasibility of the evaluation system method, we use environmental detectors to verify whether the evaluation results are correct when evaluating abnormal water quality.

(8) Online water quality evaluation

Design an online abnormal water quality evaluation system as shown in Figure 7, evaluate the water quality status in real time, extract fish movement behavior parameters in the water quality online, and use the water quality evaluation model to evaluate water quality abnormalities. At the same time, the feature parameter database and the evaluation model are updated according to the evaluation results to improve the robustness of the model. Use the evaluation results to train the SVM classifier online, that is, to train and update the SVM every certain number of video sequence image frames. Through continuous training and online updates, the water quality evaluation system is constantly adaptively changing in the time series, even under temperature and light conditions. , complex background interference and other situations, the evaluation system can still run stably. In this way, the water quality evaluation model is updated through the real-time evaluation results to improve the robustness of the evaluation system.

Claims (1)

1., based on a biological formula exception water quality evaluation system method for designing for pattern-recognition, its content comprises the steps:

(1) water quality assessment platform is built

Build water quality assessment platform, utilize computer vision technique that live body fish movement is converted into vedio data;

(2) Image semantic classification

Adopting image processing techniques to carry out pre-service to the image obtained, first bilateral filtering process is carried out to video frame images, remaining marginal information again while denoising, then dct transform being carried out to eliminate the impact of illumination to filtered image;

(3) detection and tracking of fish movement target

Due to Kalman filtering, to have track algorithm calculated amount less and can realize the advantage of real-time follow-up, therefore adopt it to follow the tracks of fish, obtains the movement locus of fish;

(4) water quality assessment index is extracted

The movement locus obtaining fish by following the tracks of institute obtain water quality assessment index and swimming rate, travelling apart from, move about acceleration, four, corner direction parameter;

(5) characteristic parameter database is set up

Pre-service is carried out to the kinematic parameter under the normal and exception water quality extracted, rejects gross error, obtain the data after normalization, set up characteristic parameter database;

(6) Water Quality Assessment Model is set up

In order to set up Water Quality Assessment Model fast, first PCA dimension-reduction treatment being carried out to the evaluation index obtained, then adopting support vector machines structure water quality anomaly evaluation model;

(7) feasibility of water quality assessment system is investigated

In order to verify the feasibility of the biological formula exception water quality evaluation system method for designing based on pattern-recognition, when carrying out water quality anomaly evaluation, environment detector is adopted to verify that whether the result of evaluation is correct;

(8) online water quality assessment

Online design exception water quality evaluation system, On-line testing water quality Mesichthyes motor behavior parameter, utilizes Water Quality Assessment Model to carry out water quality anomaly evaluation; Meanwhile, evaluation result regeneration characteristics parameter database and evaluation model is adopted, to improve the robustness of model.