JP7424112B2 - Water quality determination device - Google Patents

Description

The present invention relates to a water quality determination device.

In recent years, there has been a shortage of labor at sewage treatment plants, and there is a need for more efficient maintenance and management of sewage treatment facilities.
For this reason, consideration is being given to replacing some of the staff's work with artificial intelligence (AI).

Patent Document 1, which is an example of the conventional technology, extracts features from image data using an autoencoder and identifies normal values and outliers using a one-class classifier such as a One-Class SVM (Support Vector Machine). A technique has been disclosed for determining an abnormality.

JP 2018-5773 Publication

However, in the above-mentioned conventional technology, since the judgment is made using a single one-class classifier, it is difficult to deal with various characteristics of water quality such as color, scum, and floc, and it is difficult to detect unknown abnormal conditions. The problem was that it was also difficult.

The present invention has been made in view of the above, and an object of the present invention is to provide a technique that can determine water quality even if an unknown abnormal condition occurs.

The present invention, which solves the above-mentioned problems and achieves the object, comprises: an image data storage unit that stores image data acquired from a target for water quality determination; and a processing process performed on the image data stored in the image data storage unit. an image processing unit that outputs processed image data; and a normal-time processed image data obtained by processing image data when the water quality to be determined is in a normal state, which is trained as training data, and outputs learned parameters. Convolution is performed using a convolution AE (Auto Encoder) learning unit, a convolution AE parameter storage unit that stores the learned parameters outputted by the convolution AE learning unit, and the learned parameters stored in the convolution AE parameter storage unit. a convolution AE estimator that constructs an AE and generates and outputs restored image data to reproduce the image data processed at the time of judgment of the water quality determination target; and the processed image data at the time of judgment and the restored image. This water quality determination device includes a water quality determination section that determines water quality by comparing data.

The water quality determination device configured as described above further includes an abnormal area identifying section, wherein the water quality determining section compares the image data processed at the time of determination with the restored image data to generate difference image data, and identifies the abnormal area. It is preferable that the unit identifies the abnormal area based on the difference image data.

In the water quality determination device having the above configuration, a water quality determination data and abnormal area data storage unit that stores water quality determination data outputted by the water quality determination unit and abnormal area data outputted by the abnormal area identification unit; Preferably, the apparatus further includes a time series analysis section that performs time series analysis on the abnormal region data.

According to the present invention, the water quality can be determined even if an unknown abnormal condition occurs.

1 is a block diagram showing the configuration of a water quality determination device according to Embodiment 1. FIG. FIG. 2 is a diagram illustrating an input/output relationship between a convolution AE estimation unit and a water quality determination unit shown in FIG. 1 and an overview of convolution AE. FIG. 2 is a block diagram showing the configuration of a water quality determination device according to a second embodiment. FIG. 3 is a block diagram showing the configuration of a water quality determination device according to a third embodiment.

Embodiments of the present invention will be described below with reference to the drawings.
However, the present invention is not limited to the description of the embodiments below.

<Embodiment 1>
FIG. 1 is a block diagram showing the configuration of a water quality determination device 100 according to this embodiment.
The water quality determination device 100 shown in FIG. Equipped with.

The image data storage unit 101 receives as input image data obtained by a camera (not shown) installed in the water quality determination target, and stores the image data.
Examples of water quality determination targets include sedimentation ponds or water tanks in sewage treatment facilities and the like.
The image data stored in the image data storage section 101 is output to the image processing section 102, the convolution AE estimation section 105, and the water quality determination section 106.
The image data storage unit 101 can be realized by a recording medium such as a semiconductor memory and a magnetic disk.
This image data includes both still images and moving images composed of a plurality of consecutive still images.

The image processing unit 102 inputs image data stored in the image data storage unit 101, performs processing on the image data, and outputs processed image data.
The processing performed by the image processing unit 102 is trimming processing, normalization processing, or filter processing performed on image data so that the image data can be input to the convolution AE learning unit 103.
The image processing unit 102 can be implemented by a processor such as an MPU (Micro-Processing Unit) and a CPU (Central Processing Unit).
Note that the processed image data includes normal-time processed image data and judgment-time processed image data, which will be described later.

The convolution AE learning unit 103 receives the processed image data output from the image processing unit 102 as input, performs learning using the input image data, which is normal processed image data, as teacher data, generates learned parameters, and performs convolution AE. It is output to the parameter storage section 104.
Here, learning is performed by extracting features of the input normal-time processed image data and generating a restored image so as to reproduce the input normal-time processed image data.
The convolutional AE learning unit 103 can be implemented by a processor such as an MPU or a CPU.

Here, the learning performed by the convolutional AE learning unit 103 is performed when the water quality to be determined is in a normal state.
A normal state is a state in which the system including the water quality determination target is operating without any problems.
The convolution AE learning unit 103 uses normal-time processed image data, which is input image data that is input to the image data storage unit 101 when the water quality to be determined is in a normal state and processed by the image processing unit 102. Features are automatically extracted, restored image data is generated to reproduce the input image data, learning is performed, and learned parameters are generated.

The convolutional AE parameter storage unit 104 receives as input the learned parameters output by the convolutional AE learning unit 103, and stores the learned parameters.
The convolutional AE parameter storage unit 104 can be realized by a recording medium such as a semiconductor memory and a magnetic disk.

The convolution AE estimation unit 105 inputs the learned parameters stored in the convolution AE parameter storage unit 104 and the image data processed at the time of determination from the image processing unit 102 at the time of determination, and calculates the convolution AE using the learned parameters. is constructed to generate restored image data so as to reproduce the input judgment-processed image data.
The convolutional AE estimation unit 105 can be implemented by a processor such as an MPU or a CPU.

FIG. 2 is a diagram illustrating the input/output relationship of the convolutional AE estimation unit 105 and the water quality determination unit 106 shown in FIG. 1 and an overview of the convolutional AE estimation.
As shown in FIG. 2, the convolutional AE estimation unit 105 includes an encoder that compresses dimensions and a decoder that restores dimensions.
The convolution AE estimation unit 105 generates restored image data from input image data.
Then, the water quality determination unit 106 generates differential image data from the difference between the input image data and the restored image data.
In this way, convolutional AE does not require any special tuning in its design, and learning is performed only by inputting image data.

The convolutional AE estimation unit 105 constructs a convolutional AE using learned parameters learned from data when the water quality to be determined is in a normal state.
Therefore, if the water quality is normal when the image data is acquired for the water quality determination target, there will not be a large difference between the input processed image data and the generated restored image data.
If the water quality is not normal when the image data is acquired for the water quality determination target, that is, if it is abnormal, the restored image data will not be generated correctly at the abnormal location, and the input processed image data and restored image will not be correctly generated. There will be a difference between the two.

The water quality determination unit 106 inputs the image data processed at the time of determination, which is the input image data output by the image processing unit 102, and the restored image data output by the convolution AE estimation unit 105, and combines the input image data and the restored image data. A difference image is generated by comparing the images, water quality is determined based on the difference image, and water quality determination data indicating the water quality determination result is output.
The water quality determination unit 106 can be realized by a processor such as an MPU and a CPU.

The image data processed at the time of determination and the restored image data input to the water quality determination unit 106 are three-dimensional matrix data based on length, width, and number of channels.
The water quality determination unit 106 generates difference image data by calculating the difference between the image data processed at the time of determination and the restored image data at each pixel value.
In this difference image data, each pixel value becomes large in a portion where the difference between the input processed image data at the time of determination and the restored image is large.
That is, since the average brightness value of the difference image becomes large at an abnormal location, it is possible to evaluate the difference based on the magnitude of the average brightness value of the difference image data.
Then, the water quality determination unit 106 determines whether the water quality to be determined is normal or abnormal, using a preset threshold of the average brightness value of the difference image.
The threshold value of the average brightness value may be determined by considering the distribution of the average brightness values obtained by applying a sufficient amount of images.

As described above, according to the present embodiment, an abnormality in the water quality to be determined can be detected by performing learning using image data when the water quality to be determined is in a normal state.
According to this embodiment, an abnormality in the water quality to be determined can be detected without defining the abnormality in the water quality to be determined as a cluster in advance, and without acquiring image data at the time of the abnormality. .
Therefore, according to this embodiment, even if an unknown abnormality occurs in the water quality to be determined, the abnormality can be detected and the water quality can be determined.

<Embodiment 2>
In Embodiment 1, the water quality determination device 100 was described that can detect when an abnormality occurs in the water quality to be determined, even if the abnormality is an unknown abnormality. However, in this embodiment, only the presence or absence of an abnormality is described. Instead, a configuration that can obtain information on the position and shape of the cause of an abnormality and that can determine water quality with higher accuracy than in the first embodiment will be described.

FIG. 3 is a block diagram showing the configuration of the water quality determination device 100A according to this embodiment.
The water quality determination device 100A shown in FIG. 3 is different from the water quality determination device 100 shown in FIG. It's the same.

The water quality determination unit 106a differs from the water quality determination unit 106 in that it outputs not only water quality determination data but also generated difference image data.

The abnormal area identification unit 107 inputs the difference image data output by the water quality determination unit 106a, and identifies an area where each pixel value of the input difference image data is large, that is, an area where the average brightness value of the difference image is large, as an abnormal area. Then, abnormal area data indicating the identified abnormal area is output.
The abnormal area identification unit 107 can be realized by a processor such as an MPU and a CPU.

In addition, the abnormal area identification unit 107 first performs a binarization process on the difference image data, and repeatedly divides the binarized difference image data into a plurality of areas to identify an abnormal area.
The abnormal area identification unit 107 first divides the difference image data into a 2×2 mesh, and compares the average brightness value of the entire difference image data with the average brightness value of each divided area. It is determined whether each divided area is normal or abnormal.
Here, since the difference image data is divided into 2×2 meshes, the abnormal area identification unit 107 compares the average brightness values of the four areas of the difference image data and selects an area with a large average brightness value as the abnormal area. .
Thereafter, the abnormal region identifying unit 107 divides the abnormal region identified by the 2×2 mesh into 4×4 meshes, and similarly divides the differential image data by 2×2 meshes into 4×4 meshes. Identify abnormal areas using 4 meshes.
Thereafter, the abnormal area identification unit 107 similarly identifies an abnormal area using an 8×8 mesh.
In this way, by subdividing the mesh that divides the differential image data, it becomes possible to specify the abnormal region in detail.

Note that when the abnormal area includes particle-shaped floating substances such as scum, the abnormal area specifying unit 107 may specify the number of particles by particle analysis.
Specifically, by performing binarization processing on the difference image data, and performing smoothing filtering, morphology processing, and labeling processing on the data that has been subjected to the binarization processing, the particle positions and number of particles in the difference image data can be determined. It becomes possible to judge.
In this way, it is possible to clarify the shape, position, number, etc. of scum etc. that are the cause of the abnormality.

As explained above, according to this embodiment, in addition to the effects of Embodiment 1, it is possible not only to detect an abnormality but also to specify an abnormal region.
Therefore, according to the present embodiment, the details of the cause of the abnormality become clear, making it possible to obtain not only the presence or absence of an abnormality but also detailed information on the cause of the abnormality, making it possible to determine water quality with higher accuracy than in the first embodiment. .

<Embodiment 3>
Embodiment 1 describes a water quality determination device 100 that is capable of detecting an abnormality in the water quality to be determined, even if the abnormality is an unknown abnormality, and Embodiment 2 further describes an abnormality at a predetermined point in time. Although the water quality determination device 100A that can obtain detailed information on the cause has been described, in this embodiment, the water quality determination device 100A is more accurate than in Embodiments 1 and 2 and can obtain information on changes over time of the cause of the abnormality. A configuration that allows determination will be described.

FIG. 4 is a block diagram showing the configuration of the water quality determination device 100B according to this embodiment.
The water quality determination device 100B shown in FIG. 4 differs from the water quality determination device 100A shown in FIG. 3 in that it further includes a water quality determination data and abnormal area data storage unit 108 and a time series analysis unit 109, and the other configurations are the same. be.

The water quality determination data and abnormal area data storage unit 108 receives the water quality determination data output by the water quality determination unit 106a and the abnormal area data output by the abnormal area identifying unit 107, and stores the water quality determination data and abnormal area data. .
Note that the water quality determination data and the abnormal area data include acquisition time information at which image data was acquired for the water quality determination target.
The water quality determination data and abnormal area data storage unit 108 can be realized by a recording medium such as a semiconductor memory and a magnetic disk.

The time series analysis unit 109 inputs the water quality determination data and abnormal area data stored as time series data in the water quality determination data and abnormal area data storage unit 108, performs time series analysis based on the acquisition time information, and performs time series analysis based on the acquisition time information. Output time-series analysis data showing analysis results.
The time series analysis unit 109 performs time series analysis by performing smoothing processing such as a moving average, a Kalman filter, or a particle filter on the time series data of the water quality determination results and abnormal area identification results.
The time series analysis unit 109 can be realized by a processor such as an MPU and a CPU.

As explained above, according to this embodiment, in addition to the effects of Embodiment 2, it is possible not only to specify an abnormal region but also to obtain information on changes over time in the abnormal region.
Therefore, according to the present embodiment, it is possible to obtain information on the change over time of the cause of the abnormality, and water quality can be determined with higher accuracy than in the second embodiment.

According to the water quality determination apparatus described in Embodiments 1 to 3, specifically, the amount of suspended matter such as scum floating on the treated water that is subject to water quality determination, and the treatment that changes due to the activated sludge sedimentation rate, etc. It is possible to determine water quality by detecting abnormalities such as the color of water, but any abnormality can be detected, including but not limited to, as long as the image data shows a difference from the normal state at the time of abnormality. be able to.

100, 100A, 100B Water quality determination device 101 Image data storage unit 102 Image processing unit 103 Convolution AE learning unit 104 Convolution AE parameter storage unit 105 Convolution AE estimation unit 106, 106a Water quality determination unit 107 Abnormal area identification unit 108 Water quality determination data and abnormality Area data storage unit 109 Time series analysis unit

Claims (3)

an image data storage unit that stores image data acquired from the water quality determination target;
an image processing unit that performs processing on the image data stored in the image data storage unit and outputs processed image data;
a convolution AE learning unit that learns normal-time processed image data obtained by processing image data when the water quality to be determined is in a normal state as teacher data, and outputs learned parameters;
a convolutional AE parameter storage unit that stores the learned parameters output by the convolutional AE learning unit;
Constructing a convolutional AE using the learned parameters stored in the convolutional AE parameter storage unit, and generating and outputting restored image data so as to reproduce the judgment-processed image data at the time of judgment of the water quality judgment target. a convolutional AE estimation unit that performs
A water quality determination device comprising: a water quality determination unit that performs water quality determination by comparing the image data processed at the time of determination and the restored image data using a preset threshold of an average brightness value of a difference image . It further includes an abnormal area identification section,
The water quality determination unit compares the image data processed at the time of determination and the restored image data to generate difference image data;
The water quality determination device according to claim 1, wherein the abnormal area specifying unit specifies the abnormal area based on the differential image data , and specifies the particle position and number of particles in the abnormal area by particle analysis . a water quality determination data and abnormal area data storage unit that stores water quality determination data output by the water quality determination unit and abnormal area data output by the abnormal area identification unit;
The water quality determination device according to claim 2, further comprising a time series analysis unit that performs time series analysis on the water quality determination data and the abnormal area data.

Images