Intelligent river pollution monitoring system based on computer vision

Abstract

The invention relates to the technical field of image data processing, in particular to an intelligent river pollution monitoring system based on computer vision and is used for operating an intelligent river pollution monitoring method based on computer vision, and the system comprises a collection module which is used for setting collection parameter values and collecting a plurality of frames of river environment images in real time; the data processing module is used for defining a moving object in the river channel environment image as an image target and detecting and positioning the image target to obtain a moving vector value of the image target; the data analysis module is used for calculating a vector drift judgment quantity of the image target in combination with the moving vector value, and obtaining a predictive factor and a moving track; the monitoring module is used for analyzing the moving track noise value of the image target, determining a to-be-enhanced area in the river channel environment image, and performing enhancement processing on the to-be-enhanced area to obtain an enhancement result of the river channel environment image so as to realize intelligent monitoring of the image target; moving objects such as pollutants in the river channel environment can be efficiently recognized and processed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image data processing, in particular to a river pollution intelligent monitoring system based on computer vision. BACKGROUND

[0002] The original intention of the river pollution intelligent monitoring system is to provide a precise and efficient solution for river pollution management through real-time image acquisition, processing, recognition and intelligent decision-making. The system covers various monitoring needs such as real-time performance, environmental complexity processing capability, diversity, etc., including image acquisition and recognition, data storage communication and other key links. In the river pollution intelligent monitoring system, image enhancement processing is a crucial task. Due to the complexity and dynamics of the river environment, various challenges may be encountered during image processing, including light changes, reflections, dynamic water flow, shadows, noise and processing efficiency. In order to overcome these problems, optimization of hardware devices and enhancement of intelligent algorithms are needed, such as using HDR technology, reflection filtering, motion compensation, noise removal and edge computing techniques to achieve efficient and real-time image enhancement. This can provide high-quality image data support for subsequent pollutant recognition, thereby improving the accuracy and efficiency of river pollution management.

[0003] However, in real life, when monitoring the river environment in real time, pollutants usually float in a flowing state. During the process of pollutant monitoring by recognition algorithm, due to the complexity of the environment, under the condition of natural light change, especially in weak light conditions, when the pollutants float into the shadow area formed by the riverbank vegetation or bridge structure, the details of the pollutants may be covered or blurred. In this case, if global algorithm enhancement is used, although the visibility of the image can be improved, it will greatly affect the real-time performance of recognition, resulting in the inability to identify the classification result of the pollutants in time, which will directly affect the effect of river pollution monitoring. SUMMARY

[0004] In order to solve the technical problem that the existing method has low light, the shadow area formed by the riverbank vegetation or bridge, etc. in the monitoring process of the complex river environment, which causes the pollutants in the river to be covered or blurred, affecting the real-time performance of recognition, and resulting in the inability to accurately monitor, the purpose of the present application is to provide a river pollution intelligent monitoring system based on computer vision, which is used to run a river pollution intelligent monitoring method based on computer vision. The system comprises:

[0005] The acquisition module is configured to set the acquisition parameter value and acquire a plurality of frames of river environment images in real time.

[0006] The data processing module is configured to define a moving object in the river environment image as an image target, detect and locate the image target, and obtain a moving vector value of the image target.

[0007] The data analysis module is configured to calculate a vector drift determination quantity of the image target in combination with the moving vector value, obtain a prediction factor and a moving track, and determine the moving track noise value of the image target.

[0008] The monitoring module is configured to analyze the moving track noise value of the image target, determine a region to be enhanced in the river environment image, perform enhancement processing on the region to be enhanced, obtain an enhancement result of the river environment image, and realize intelligent monitoring of the image target.

[0009] Preferably, the moving object in the river environment image is defined as the image target, the image target is detected and located, and the moving vector value of the image target is obtained, including:

[0010] The river environment image is subjected to background modeling and foreground segmentation, and a difference result value of the image target in each frame of the river environment image is obtained to generate a difference set sequence.

[0011] A deep learning detection algorithm is used to perform real-time positioning on the image target, and a positioning mark box is generated based on the image target, the coordinates of the image target under different frames are obtained through the positioning mark box, and a coordinate set sequence is integrated and generated.

[0012] The moving distance and angle value of the image target between adjacent frames are obtained according to the coordinate set sequence, the moving distance average and angle value average of a plurality of adjacent frames are determined respectively, the drift speed value of the image target is calculated, and the moving vector value of the image target is determined in combination with the angle value average.

[0013] The difference set sequence and the moving vector value of the image target of a plurality of frames are transmitted to the data analysis module.

[0014] Preferably, the drift speed value of the image target is calculated, and the corresponding calculation formula is:

[0015]

[0016] wherein t represents the river environment image corresponding to the t frame; v t represents the drift speed value of the image target in the t frame. represents the moving distance average; t0 represents a fixed time between adjacent frames.

[0017] Preferably, the moving vector value of the image target is determined in combination with the angle value average, and the corresponding calculation formula is:

[0018]

[0019] wherein, A t represents a moving vector value of the image target; v t A t represents a drift velocity value of the image target in the t frame; A t represents an average value of the angle value.

[0020] Preferably, the vector drift determination quantity of the image target is calculated in combination with the moving vector value, the prediction factor and the moving track are obtained, including:

[0021] The difference result value in the difference set sequence is sequentially recorded as a vector drift determination value weight, and the vector drift determination quantity of the image target is calculated in combination with the moving vector value;

[0022] A numerical range is set, the image target to be obtained for the prediction factor and the moving track is determined according to the vector drift determination quantity of the image target, and the screened vector drift determination quantity is recorded as the prediction factor;

[0023] The moving track of the image target is obtained through ARIMA model based on the prediction factor;

[0024] The moving track of the image target is transmitted to the monitoring module.

[0025] Preferably, the vector drift determination quantity of the image target is calculated, and the corresponding calculation formula is:

[0026]

[0027] Wherein, A t A t represents the vector drift determination quantity of the image target in the t frame; e t A t represents the difference result value of the image target in the t frame; A t represents a moving vector value of the image target.

[0028] Preferably, the prediction analysis is carried out through ARIMA model based on the prediction factor, and the corresponding calculation formula is:

[0029] A m+i = φ0+ φ1A m-1 +…+ φ p φ m-p + ∈ t

[0030] Wherein, A m+i A m+i represents the m+i prediction result, and the prediction result includes the moving speed and direction of the image target; φ0, φ1... φ p A t represents the autoregressive coefficient; A m-1 ... A m-p A t represents the prediction factor; m-1... m-p represents the serial number of the prediction factor; ∈ t A t represents the error term.

[0031] Preferably, the moving track noise value of the image target is analyzed, the region to be enhanced in the river environment image is determined, and the region to be enhanced is enhanced to obtain an enhanced result of the river environment image, so that intelligent monitoring of the image target is realized, including:

[0032] The moving track noise value of the image target is analyzed, a judgment value range is set, and the region corresponding to the moving vector value in the judgment value range is determined as the region to be enhanced in the river environment image;

[0033] The enhancement scale size of the region to be enhanced is obtained in combination with the moving track noise value, linear enhancement is performed, and enhancement processing is performed according to the enhancement scale size, so that intelligent monitoring of the image target is realized.

[0034] Preferably, the moving track noise value of the image target is analyzed, and the corresponding calculation formula is:

[0035]

[0036] Wherein, No m+i represents the moving track noise value of the m+i th prediction result; g v represents the pixel value of the v th pixel point in the positioning mark box corresponding to the m+i th prediction result; Q represents the total Q pixel point pixel values in the positioning mark box corresponding to the m+i th prediction result; En represents the entropy value of the pixel point pixel value; morm() represents the maximum and minimum value normalization function.

[0037] Preferably, the enhancement scale size of the region to be enhanced is obtained in combination with the moving track noise value, linear enhancement is performed, and enhancement processing is performed according to the enhancement scale size, including:

[0038] The enhancement scale size of the region to be enhanced is obtained in combination with the moving track noise value, and the corresponding calculation formula is:

[0039] b=1+No m+i

[0040] Wherein, b represents the enhancement scale size of the region to be enhanced; No m+i represents the moving track noise value of the m+i th prediction result;

[0041] Linear enhancement is performed, and the corresponding calculation formula for enhancement processing according to the enhancement scale size is:

[0042] y=bx+c

[0043] Wherein, y represents the output pixel value after enhancement; x represents the input pixel value before enhancement; b represents the enhancement scale size of the region to be enhanced; and c represents the intercept of linear enhancement.

[0044] The present application has the following advantages:

[0045] Through real-time acquisition of riverway environment image, data support is provided for subsequent identification enhancement; through mobile detection and positioning of the image target, target mobile vector value is obtained, and the target vector drift determination quantity is combined to determine the prediction factor and the moving track, so as to enhance the image target, without losing the mobile characteristics of the image target in the technology of ensuring real-time enhancement, while reducing the image data calculation amount, the visibility of the image target can be improved; the moving track noise value is analyzed, the real-time positioning result is combined, the enhancement scale size of the to-be-enhanced area is determined, and then the enhancement result of the entire riverway environment image is obtained, so as to achieve the purpose of intelligent monitoring, and the mobile objects such as pollutants in the riverway environment are efficiently identified and processed. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0047] Figure 1 a schematic block diagram of the computer vision-based riverway pollution intelligent monitoring system provided by an embodiment of the present application;

[0048] Figure 2 a schematic diagram of the positioning mark of the computer vision-based riverway pollution intelligent monitoring system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined application purpose, the computer vision-based riverway pollution intelligent monitoring system according to the present application, its specific implementation, structure, features and effects are described in detail as follows in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0051] The specific scheme of the computer vision-based riverway pollution intelligent monitoring system provided by the present application is described in detail below in combination with the drawings.

[0052] When real-time monitoring is performed on a complex river environment, the moving objects in the river environment, including pollutants and oil stains, are mostly in a flowing state and constantly float. During natural light changes, when the moving objects float into a shadow area formed by riverbank vegetation or a bridge, the details of the moving objects are covered or blurred under low light conditions. If global algorithm enhancement is performed on the river environment image at this time, the real-time performance of moving object identification will be greatly affected, the moving objects cannot be identified for classification, and the monitoring effect of the river environment is affected. In an embodiment of the present application, a river pollution intelligent monitoring system based on computer vision is provided. The system acquires river environment images in real time, detects and locates the moving objects in the images, obtains target moving vector values, determines a prediction factor and a moving track in combination with a target vector drift determination quantity, analyzes a moving track noise value, determines the enhancement scale of the region to be enhanced in combination with a real-time positioning result, and then obtains the enhancement result of the entire river environment image, thereby achieving intelligent monitoring of the moving objects. The system is used to implement a river pollution intelligent monitoring method based on computer vision. In essence, the system is a software system composed of various modules that realize corresponding functions. The specific steps of the modules in the system are described in detail.

[0053] Referring to Figure 1 , a schematic block diagram of a river pollution intelligent monitoring system based on computer vision is shown. The system is used to implement a river pollution intelligent monitoring method based on computer vision. The system includes:

[0054] An acquisition module is configured to set acquisition parameter values and acquire a plurality of frames of river environment images in real time.

[0055] A data processing module is configured to define moving objects in the river environment images as image targets, detect and locate the image targets, and obtain moving vector values of the image targets.

[0056] A data analysis module is configured to calculate a vector drift determination quantity of the image targets in combination with the moving vector values, and obtain a prediction factor and a moving track.

[0057] A monitoring module is configured to analyze a moving track noise value of the image targets, determine a region to be enhanced in the river environment image, perform enhancement processing on the region to be enhanced, obtain an enhancement result of the river environment image, and achieve intelligent monitoring of the image targets.

[0058] It can be explained that the river pollution intelligent monitoring system based on computer vision stores program data. When the program data is running, the river pollution intelligent monitoring method based on computer vision is implemented. The system mainly monitors river water quality and moving objects such as pollution sources in a high-efficiency and real-time manner, which is conducive to forming a complete pollution prevention and management solution, timely warning, and taking preventive measures to ensure the health of the river ecological environment.

[0059] It is explained that the acquisition module is used to obtain a plurality of frames of river environment images to provide basic data support for the subsequent data processing module; the data processing module is used to process the river environment images to obtain the moving vector value of each image target, so as to ensure the accuracy and timeliness of the image data required by the user to be monitored; the data analysis module is used to determine the prediction factor and the moving track of the image target to determine the image target that needs to be enhanced; the monitoring module is used to analyze the moving track noise value of the image target, combined with the aforementioned real-time positioning result, to realize intelligent monitoring and management, to timely discover and report any abnormal situation, and to ensure the pollution treatment of the river environment.

[0060] As an optional implementation, the monitoring module can also perform coordination and early warning processing on the region to be enhanced to remind the river environment maintenance personnel to make subsequent processing, thereby enhancing the intelligence of the entire system.

[0061] It can be understood that the modules of the river pollution intelligent monitoring system based on computer vision need to run the river pollution intelligent monitoring method based on computer vision when in operation, so whether the acquisition module, the data processing module, the data analysis module and the monitoring module are integrated or different hardware is configured to produce similar functions to the effects achieved by the present application, all belong to the protection scope of the present application.

[0062] It can be explained that the steps implemented by the acquisition module include:

[0063] A high-resolution monitoring camera is used to set the acquisition parameter values of frame rate, resolution, exposure time and ISO sensitivity in sequence, to real-time collect a plurality of frames of river environment images, and to transmit all the river environment images to the data processing module.

[0064] It is explained that the high-resolution camera refers to a camera that can provide higher image resolution, such as 1080p, 4K or higher resolution, which can capture more pixel information and provide clearer and more delicate river environment images, capture more details, and effectively improve the real-time collection effect; all the river environment images are transmitted to the data processing module to support edge computing and cloud processing collaborative work, and to optimize real-time performance and processing efficiency.

[0065] As an optional implementation, in the embodiment, the frame rate represents the frequency of the continuous appearance of the bitmap image in the frame on the display, which is set to 25-30 frames per second to ensure smooth picture transmission and adapt to the moving object monitoring requirement; the resolution refers to the number of pixel points contained in a unit inch, which is set to 1080p (1920x1080) to balance the image definition and transmission real-time performance; the exposure time refers to the time for which the switch is opened to project light onto the light-sensitive surface of the camera, and a typical value range of 1 / 30 second-1 / 100 second is set, and the exposure time is automatically lengthened for the low-light area in the riverway environment image to avoid excessive exposure and water surface reflection; the ISO (International Organization for Standardization) sensitivity is used to measure the sensitivity of the light-sensitive element to light, which is set to 100-800, i.e., low sensitivity, to adapt to the light changes in the riverway environment, and the ISO sensitivity is increased to improve the brightness when in the shadow area formed by the riverbank vegetation or bridge, thereby avoiding the increase of noise in the riverway environment image.

[0066] Further, the data processing module is used to implement the following steps.

[0067] Step S21: The background modeling and foreground segmentation are performed on the riverway environment image to obtain a difference result value of the image target in each frame of the riverway environment image, and a difference set sequence is generated.

[0068] It is explained that the background modeling refers to establishing a stable background image to compare with the image of the current frame in the subsequent frames to detect the foreground, i.e., the moving object, and the Gaussian mixture model or the mean background modeling method is usually used, in which the Gaussian mixture model is to establish a plurality of Gaussian distribution models for the color value, RGB or grayscale value of each pixel to represent the change of the background, the color of each pixel is compared with the color of the historical frame based on the color of the current frame, if matched, the model is updated, if not matched, it is considered as the foreground and marked as the foreground pixel; the mean background modeling is to calculate the historical average value of each pixel position, i.e., to represent the background by the mean value of the pixel value in each frame image, if the difference between the current frame and the background image is large, it is considered as the foreground.

[0069] The foreground segmentation is to separate the part other than the background from each frame, that is, to identify the moving objects in the image, so as to effectively extract the dynamic target from the river environment image; in the embodiment, the background difference method is used to extract the moving objects in the river environment image in several frames to obtain the difference result value of the image target, wherein the image target is mainly the solid floating object, oil stain, chemical substance and other pollutants polluting the river environment, that is, the inter-frame difference method is used, which is simple and not easily affected by the environmental light, that is, the pixel values of two images of adjacent or several frames apart in the river environment image are subtracted, and each river environment image after subtraction is thresholded to extract the image target, the difference between the image target and the background in each frame of the river environment image is the difference result value, and then the difference set sequence is generated.

[0070] Specifically, the image target of each frame has a difference result value e t , which represents the difference result of different frames and the previous frame, and the difference set sequence {e1, e2, …, e t} is generated after normalization, the larger the difference result value is, the clearer the image target is, and the image target in the image is characterized by the change of pixel coordinate position.

[0071] Optionally, in the specific practice process, the background model is updated adaptively to filter out the water flow, ripple and other dynamic noise, so as to prevent the accuracy of subsequent image target positioning from being affected by such noise, wherein the median filter or bilateral filter is used to reduce the small-scale noise caused by the ripple and the like.

[0072] Please refer to Figure 2 , which shows the positioning mark of the river pollution intelligent monitoring system based on computer vision provided by an embodiment of the application.

[0073] Step S22: using a deep learning detection algorithm to perform real-time positioning on the image target, and generating a positioning mark frame based on the image target, obtaining the coordinates of the image target under different frames through the positioning mark frame, and integrating to generate a coordinate set sequence.

[0074] It can be explained that in the embodiment, YOLO is used to perform real-time positioning on the image target in the river environment image, which can detect the image target at low resolution and reduce the calculation cost, wherein YOLO (You Only Look Once) is a single detection algorithm, which converts the image target detection problem into a regression problem, the core idea is to divide the input river environment image into a grid, and each grid unit is responsible for predicting the object it contains, for each grid unit, the YOLO algorithm predicts a confidence score, which represents the probability that the unit contains the image target, and a set of bounding box coordinates, which represents the position of the image target in each frame of the river environment image; the YOLO algorithm is fast and can process the river environment image in real time.

[0075] It can be understood that the moving object in the river environment will greatly reduce the accuracy of its identification when it drifts to a low-light area, i.e. a shadow area formed by riverbank vegetation or a bridge, due to its mobility, so the foregoing scenario needs to be enhanced to improve the real-time performance of the processing effect and prevent other areas of the river environment from causing poor visual effects due to excessive enhancement. In order to enhance specifically, the drift characteristics and path of the image target need to be analyzed and obtained, and then the corresponding specific area is enhanced.

[0076] Specifically, through the positioning results of different frames obtained in the foregoing, the positioning results represent that different image targets with different positioning needs exist in different frames, which are updated over time, and a positioning mark box is generated based on the image target to obtain the geometric center coordinate value of the positioning mark box in each frame, denoted as r t (u) = {i, j}, which represents the center geometric coordinate corresponding to the positioning mark box of the u-th image target in the t-th frame, i and j respectively represent the coordinate values corresponding to the center of the lower left corner of each frame of river environment image in the Cartesian coordinate system, after the geometric center coordinate values of the image targets in all frames are obtained, a coordinate set sequence is integrated and generated, denoted as {r1, r2, …, r t}, wherein the geometric center coordinate values reflect the motion direction and speed of the image target in different frames.

[0077] Step S23: According to the coordinate set sequence, the moving distance and angle value of the image target in adjacent frames are obtained, the moving distance average and angle value average of a plurality of adjacent frames are determined respectively, the drift speed value of the image target is calculated, and the moving vector value of the image target is determined in combination with the angle value average.

[0078] Further, in step S23, the drift speed value of the image target is calculated, and the corresponding calculation formula is:

[0079]

[0080] Wherein, t represents the river environment image corresponding to the t-th frame; v t represents the drift speed value of the image target in the t-th frame; represents the moving distance average; t0 represents the fixed time between adjacent frames.

[0081] It is explained that the drift speed value of the image target is determined based on a short time, wherein the short time refers to a differential time, which represents a change or measurement occurring in a relatively short time, i.e. the time difference between two frames of river environment images; and t0 represents the fixed time between adjacent frames, i.e. the time interval displayed by each frame is fixed, which affects the smoothness of the entire river environment image display.

[0082] Further, in step S23, the moving vector value of the image target is determined in combination with the average angle value, and the corresponding calculation formula is:

[0083]

[0084] wherein, v represents the moving vector value of the image target; v t represents the drift speed value of the image target in the t frame; represents the average angle value.

[0085] It can be explained that the moving vector value is used to describe the moving direction and moving speed of the image target between consecutive frames, that is, the displacement amount from a certain pixel or pixel block in the current frame to the corresponding pixel or pixel block in the next frame, including the horizontal direction component and the vertical direction component.

[0086] It is explained that the moving vector value is determined to analyze the moving track of the image target, so as to perform prediction processing on the corresponding moving vector values in different frames; after obtaining the moving track of the image target, even if the image target moves to a region that is difficult to observe and the positioning result is poor, the real-time large-scale enhancement processing on the track can be performed according to the moving track result, so as to provide sufficient pre-processing for subsequent recognition.

[0087] Step S24: transmitting the difference set sequence and the moving vector values of the image target in several frames to the data analysis module.

[0088] It can be understood that determining the vector drift determination value means determining the accuracy or visibility of the positioning result of the image target with the change of time based on several frames; for the part with lower visibility, it is determined that it has drifted to a region that needs to be enhanced and does not participate in the prediction of the track to prevent prediction error; that is, by analyzing the difference result value of the image target in different frames, the difference result value is used as the vector drift determination value weight; the larger the difference result value, the clearer the image target, and the higher the necessity of the corresponding vector drift determination value participating in the prediction of the moving track.

[0089] Further, the steps implemented by the data analysis module include:

[0090] Step S31: taking the difference result value in the difference set sequence as the vector drift determination value weight in turn, and calculating the vector drift determination amount of the image target in combination with the moving vector value.

[0091] Further, in step S31, the vector drift determination amount of the image target is calculated, and the corresponding calculation formula is:

[0092]

[0093] wherein, A ta vector drift determination quantity representing the t-th frame image target; e t a difference result value representing the t-th frame image target; a movement vector value representing the image target.

[0094] It is explained that e t The greater or closer to 1, the greater the difference between the image target and the background of the river environment image, the more obvious the morphological characteristics of the image target, that is, the better the visual effect of the image target; at this time, the vector drift determination quantity corresponding to the image target is involved in the prediction of the movement trajectory, and the movement trajectory credibility obtained is higher, that is, the morphological characteristics are more obvious, at this time, the vector drift determination value weight of this type of image target is higher, and the vector drift determination quantity is higher.

[0095] It can be explained that the difference result value e t It refers to the difference between the image target and the background in each frame of the river environment image, which indicates whether the image target has changed significantly, and generates a difference set sequence {e1, e2, …, e t} after normalization processing, that is, it can be explained that the difference result value e t The value range of e t The greater or closer to 1, the greater the difference between the image target and the background in the river environment image, the more significant the change between them.

[0096] Step S32: Set a screening numerical range, determine the image target of the prediction factor and the movement trajectory to be obtained according to the vector drift determination quantity of the image target, and record the screened vector drift determination quantity as the prediction factor.

[0097] It can be explained that in this embodiment, the screening numerical range is set to [0.8, 1], and the vector drift determination quantity is normalized after taking the modulus, and the image target in the screening numerical range after normalization processing is taken as the vector result participating in the calculation of the movement trajectory, and the vector drift determination quantity corresponding thereto is recorded as the prediction factor; wherein the prediction factor refers to an input variable used to predict the image target variable.

[0098] Step S33: Perform prediction analysis based on the prediction factor through the ARIMA model to obtain the movement trajectory of the image target.

[0099] For better illustration, the ARIMA (AutoRegressive Integrated Moving Average) model is a classic statistical model for time series prediction, which combines the ideas of autoregression (AR) and moving average (MA) to analyze and predict the future trend of sequence data; including the autoregressive (AR, AutoRegressive) part, which represents the relationship between the current value and the past value of the sequence data; the difference (I, Integrated) part makes the non-stationary sequence data become stationary; the moving average (MA, Moving Average) part represents the relationship between the current value and the past error term.

[0100] Further, in step S33, the prediction analysis is performed based on the prediction factor by the ARIMA model, and the corresponding calculation formula is:

[0101] A m+i = φ0+ φ1A m-1 +…+ φ p A m-p +∈ t

[0102] Where, A m+i represents the m+i prediction result, the prediction result includes the moving speed and direction of the image target; φ0, φ1... φ p represents the autoregressive coefficient; A m-1 ...A m-p represents the prediction factor; m-1... m-p represents the serial number of the prediction factor; ∈ t represents the error term.

[0103] It can be explained that the prediction result includes the moving speed and direction of the image target to display the coordinate position of the image target; based on the foregoing, it can be known that the prediction factor is a vector result, which is introduced into the ARIMA model for appropriate expansion to process multi-dimensional data, and then the position of the current analyzed image target is obtained.

[0104] It is explained that the image target moves to different areas with different enhancement needs, and each moving object has a predicted moving track; when the image target moves to some special areas, such as the shadow area formed by the riverbank vegetation or the bridge, etc., at this time, it will cause the details of the image target to be covered or blurred, so the necessity and scale of enhancement in this case are higher, and the enhanced area will appear on the predicted moving track of the image target; therefore, it is necessary to determine which areas the image target is placed in, and which areas need to be enhanced in real time.

[0105] Step S34: transmitting the moving track of the image target to the monitoring module.

[0106] It can be understood that the image target is in the shadow area formed by the riverbank vegetation or the bridge, etc., which is shown as the local gray value reduction in the river environment image, the contrast between the image target is reduced, and the complexity is increased, so when the image target is in the environment, the corresponding moving track noise value is higher, and the positioning result is combined to perform real-time enhancement processing.

[0107] Further, the monitoring module is used to realize the steps, including:

[0108] Step S41: analyze the moving track noise value of the image target, set a judgment value range, and determine the region corresponding to the moving vector value in the judgment value range as the river environment image to be enhanced.

[0109] It can be explained that the moving track noise value is to show whether the river environment image is black and chaotic; the moving track refers to the contour or boundary of the image target in each frame of the river environment image being positioned, and according to the change of the continuous frames, the displacement and moving direction of the image target are reflected; and the source of the noise is usually shown as the quality problem of the river environment image, such as illumination change, shadow, reflection, etc., which may cause inaccurate identification of the image target; or environmental interference, such as wind, object shielding, equipment error, etc.; which is shown as the instantaneous deviation or irregular fluctuation of the image target in the moving track.

[0110] Further, in step S41, the moving track noise value of the image target is analyzed, and the corresponding calculation formula is:

[0111]

[0112] Wherein, No m+i represents the moving track noise value of the m+i th prediction result; g v represents the pixel value of the v th pixel point in the positioning mark box corresponding to the m+i th prediction result; Q represents the total Q pixel point pixel values in the positioning mark box corresponding to the m+i th prediction result; En represents the entropy value of the pixel point pixel value; norm() represents the maximum and minimum value normalization function.

[0113] It is explained that the higher the entropy value is, the more chaotic the pixel value distribution near the image target is, the higher the complexity in the shadow area formed by the riverbank vegetation or the bridge, etc. is, the lower the gray mean value is, the lower the overall visibility is, the lower the contrast of the image target is, the worse the identification effect is, and the local gray value of the shadow area in the river environment image is also reduced.

[0114] As an optional implementation, in the embodiment, the judgment value range is [0.9, 1].

[0115] It can be explained that when the moving track noise amount of the prediction result is in the judgment value range, it is considered that the moving track noise amount of the corresponding image target is high, and at this time, the positioning mark frame corresponding to the image target needs to be enhanced, that is, the enhancement processing is performed on the to-be-enhanced region.

[0116] Step S42: obtain the enhancement scale size of the to-be-enhanced region in combination with the moving track noise value, perform enhancement processing according to the enhancement scale size through linear enhancement, and realize intelligent monitoring of the image target.

[0117] Further, in step S42, the enhancement scale size of the to-be-enhanced region is obtained in combination with the moving track noise value, and the enhancement processing is performed according to the enhancement scale size through linear enhancement, including:

[0118] The corresponding calculation formula for obtaining the enhancement scale size of the to-be-enhanced region in combination with the moving track noise value is:

[0119] b=1+No m+i

[0120] Wherein, b represents the enhancement scale size of the to-be-enhanced region; No m+i represents the moving track noise amount of the m+i th prediction result;

[0121] The corresponding calculation formula for performing enhancement processing according to the enhancement scale size through linear enhancement is:

[0122] y=bx+c

[0123] Wherein, y represents the output pixel value after enhancement; x represents the input pixel value before enhancement; b represents the enhancement scale size of the to-be-enhanced region; c represents the intercept of linear enhancement.

[0124] It is explained that the enhancement processing according to the enhancement scale size, that is, obtaining the enhancement result of the monitored river environment image, ensures the visibility of the main region in the river environment, avoids other regions being too enhanced, and affects the poor recognizable effect of the pollutants, and ensures the accuracy and effectiveness of the environmental monitoring; wherein, c represents the intercept of linear enhancement, which is used to control the brightness to adapt to different river environment light conditions, and in actual application, a suitable intercept is selected to achieve the best visual effect.

[0125] Optionally, the enhancement processing is performed based on the enhancement scale size to obtain an enhancement result. The identification and monitoring of the mobile object such as the pollutant are performed by using the recognition network. Specifically, first, the river environment image is preprocessed to improve the image quality and adjust the size of the river environment image, i.e., to fix the size of the river environment image to the size required by the network. Then, the river environment image is normalized to improve the stability of model training or reasoning and to improve the processing efficiency. Then, the convolutional neural network (CNN) is used to extract features in the river environment image layer by layer, such as the shape, texture, color, and the like of the mobile object such as the pollutant. The extracted features are processed by a classifier such as a Softmax classifier to output the probability of each pollutant category. A target detection algorithm such as YOLO (You Only Look Once) is used to output the position and category label of the pollutant. Finally, the result is output, i.e., the pollutant category, position, and confidence are displayed or stored for monitoring or subsequent processing.

[0126] It can be understood that the river environment image is collected in real time to provide data support for subsequent identification and enhancement. The movement detection and positioning of the image target are performed to obtain a target movement vector value. The target vector drift determination quantity is combined to determine a prediction factor and a movement trajectory to enhance the image target. In the technology of ensuring real-time enhancement, the movement characteristics of the image target are not lost, the image data calculation amount is reduced, and the visibility of the image target can be improved. The movement trajectory noise value is analyzed, and the real-time positioning result is combined to determine the enhancement scale size of the to-be-enhanced region. Then, the enhancement result of the entire river environment image is obtained to achieve the purpose of intelligent monitoring, so as to efficiently identify and process the mobile object such as the pollutant in the river environment.

[0127] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0128] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment mainly describes the differences from other embodiments.

Claims

1. A computer vision-based intelligent monitoring system for river pollution, characterized in that, The system is used to implement a computer vision-based intelligent monitoring method for river pollution, and includes: The acquisition module is used to: set acquisition parameter values ​​and acquire several frames of river environment images in real time; The data processing module is used to: define moving objects in the river environment image as image targets, detect and locate the image targets, and obtain the movement vector value of the image targets; The data analysis module is used to: calculate the vector drift determination of image targets by combining the moving vector values, and obtain the prediction factors and movement trajectories; The monitoring module is used to: analyze the noise value of the movement trajectory of the image target, determine the area to be enhanced in the river environment image, and perform enhancement processing on the area to be enhanced to obtain the enhancement result of the river environment image, thereby realizing intelligent monitoring of the image target.

2. The computer vision based intelligent monitoring system of river pollution as claimed in claim 1 wherein, Moving objects in the river environment image are defined as image targets. These targets are detected and located to obtain their motion vector values, including: Background modeling and foreground segmentation are performed on the river environment images to obtain the difference result values ​​of the image targets in each frame of the river environment image, and a difference set sequence is generated. A deep learning detection algorithm is used to locate targets in images in real time, and a location marker box is generated based on the target. The coordinates of the target in different frames are obtained through the location marker box, and the coordinate set sequence is generated by integrating them. The moving distance and angle values ​​of the image target in adjacent frames are obtained based on the coordinate set sequence. The average moving distance and average angle values ​​of several adjacent frames are determined respectively. The drift velocity value of the image target is calculated. The moving vector value of the image target is determined by combining the average angle value. The difference set sequence and the motion vector values ​​of targets in several frames of image data are transmitted to the data analysis module.

3. The computer vision based intelligent monitoring system of river pollution as claimed in claim 2, wherein, The formula for calculating the drift velocity of the image target is as follows: Wherein, t represents the riverway environment image corresponding to t frame; v t represents the drift speed value of the image target in t frame; represents the moving distance average; t0 represents the fixed time between adjacent frames.

4. The computer vision based intelligent monitoring system of river pollution as claimed in claim 3, wherein, The motion vector value of the image target is determined by combining the average angle value. The corresponding calculation formula is as follows: wherein, represents a moving vector value of the image target; v t represents a drift velocity value of the image target in the t frame; represents an average value of the angle value.

5. The computer vision based intelligent monitoring system of river pollution as claimed in claim 2, wherein, The vector drift determination of the image target is calculated by combining the moving vector value, and the prediction factor and movement trajectory are obtained, including: The difference result values ​​in the difference set sequence are recorded as vector drift determination weights in turn, and the vector drift determination quantity of the image target is calculated by combining the moving vector value. Set the range of filtering values, determine the image targets whose prediction factors and movement trajectories are to be obtained based on the vector drift judgment value of the image targets, and record the filtered vector drift judgment value as the prediction factor. The movement trajectory of the image target is obtained by predicting and analyzing the predictive factors using the ARIMA model. The movement trajectory of the image target is transmitted to the monitoring module.

6. The computer vision based intelligent monitoring system of river pollution as claimed in claim 5 wherein, The vector drift determination of the image target is calculated using the following formula: wherein A t represents a vector drift determination quantity of the t-th intra-frame image target; e t represents a difference result value of the t-th intra-frame image target; represents a moving vector value of the image target.

7. The computer vision based intelligent monitoring system of river pollution as claimed in claim 5 wherein, The ARIMA model is used for predictive analysis based on predictor factors, and the corresponding calculation formula is as follows: A m+i = φ0+ φ1A m-1 +…+φ p A m-p +∈ t wherein A m+i represents the m+1th prediction result, the prediction result including the moving speed and direction of the image target; φ0, φ1...φ p represents an autoregressive coefficient; A m-1 ...A m-p represents a prediction factor; m-1...m-p represents the serial number of the prediction factor; ∈ t represents an error term.

8. The computer vision based intelligent monitoring system of river pollution as claimed in claim 7, wherein, Analyzing the noise values ​​of the moving trajectories of image targets, identifying areas in the river environment image that need enhancement, and performing enhancement processing on these areas to obtain the enhanced river environment image, thereby achieving intelligent monitoring of image targets, including: Analyze the noise value of the moving trajectory of the target in the image, set the judgment value range, and determine the area corresponding to the moving vector value that is within the judgment value range as the area to be enhanced in the river environment image; The moving track noise value is combined to obtain the enhancement scale size of the region to be enhanced, linear enhancement is performed, and enhancement processing is performed according to the enhancement scale size, so that intelligent monitoring of the image target is realized.

9. The computer vision based intelligent monitoring system of river pollution as claimed in claim 8, wherein, The moving track noise value of the image target is analyzed, and the corresponding calculation formula is: wherein No m+i represents the moving track noise amount of the m+i-th prediction result; g v represents the pixel value of the v-th pixel point within the positioning mark frame corresponding to the m+i-th prediction result; Q represents the pixel values of the total Q pixel points within the positioning mark frame corresponding to the m+i-th prediction result; En represents the entropy value of the pixel value; and norm() represents the maximum-minimum value normalization function.

10. The computer vision based intelligent monitoring system of river pollution as claimed in claim 8, wherein, The moving track noise value is combined to obtain the enhancement scale size of the region to be enhanced, linear enhancement is performed, and enhancement processing is performed according to the enhancement scale size, including: The moving track noise value is combined to obtain the enhancement scale size of the region to be enhanced, and the corresponding calculation formula is: b = 1 + No m+i wherein b represents the enhancement scale size of the region to be enhanced; No m+i represents the moving trajectory noise amount of the m+ith prediction result; Linear enhancement is performed, and enhancement processing is performed according to the enhancement scale size, and the corresponding calculation formula is: y = bx + c Wherein, y represents the output pixel value after enhancement; x represents the input pixel value before enhancement; b represents the enhancement scale size of the region to be enhanced; and c represents the intercept of linear enhancement.

Images