CN117372942A - Reservoir floater automatic identification method based on improved SegNet model - Google Patents
Reservoir floater automatic identification method based on improved SegNet model Download PDFInfo
- Publication number
- CN117372942A CN117372942A CN202310069796.5A CN202310069796A CN117372942A CN 117372942 A CN117372942 A CN 117372942A CN 202310069796 A CN202310069796 A CN 202310069796A CN 117372942 A CN117372942 A CN 117372942A
- Authority
- CN
- China
- Prior art keywords
- reservoir
- model
- image
- point
- floaters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 23
- 208000034699 Vitreous floaters Diseases 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 20
- 238000002372 labelling Methods 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000011176 pooling Methods 0.000 claims abstract description 5
- 239000003086 colorant Substances 0.000 claims abstract description 4
- 238000012795 verification Methods 0.000 claims description 12
- 230000009467 reduction Effects 0.000 claims description 10
- 230000011218 segmentation Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 6
- 238000011426 transformation method Methods 0.000 claims description 6
- 241000251468 Actinopterygii Species 0.000 claims description 3
- 241000196324 Embryophyta Species 0.000 claims description 3
- 238000005286 illumination Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 10
- 238000012544 monitoring process Methods 0.000 description 10
- 238000007726 management method Methods 0.000 description 7
- 238000013135 deep learning Methods 0.000 description 3
- 238000007689 inspection Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The invention discloses a reservoir floater identification method based on an improved SegNet model, which comprises the following steps: acquiring a reservoir float data set by using an unmanned aerial vehicle; performing data preprocessing operation on the acquired data set; labeling the data set and dividing the data set into a training set and a testing set; training and improving a SegNet algorithm model, providing a combination of maximum pooling index and residual error connection, more comprehensively retaining a large amount of detail information contained in the multi-scale image, retaining the characteristics of colors, textures, boundaries and the like of the original image to the greatest extent, and improving the identification precision of reservoir floaters; detecting and identifying reservoir floaters; and pushing the reservoir image recognition result to relevant responsible persons and management staff. The beneficial effects of the invention are as follows: the improved SegNet algorithm model is adopted to detect and identify reservoir floaters, the training parameter quantity is moderate, the detection precision is good, and the automatic high-precision detection and identification of reservoir water surface floaters are realized.
Description
Technical Field
The invention belongs to the field of image analysis and identification, and particularly relates to an automatic reservoir floater identification method based on an improved SegNet model.
Background
A large amount of floaters in the rainwater period flow into the reservoir area along with flood, so that not only is the water quality polluted, but also the flood discharge safety of spillways and water delivery holes is affected. The large-area floaters bring great hidden trouble to the safe operation of the reservoir in the flood season. At present, the floating objects and water quality change conditions of reservoir operation management mainly depend on manual field inspection, and the manual cost is high and the inspection efficiency is low. In recent 20 years, image monitoring and video monitoring systems are increasingly widely built in reservoirs, but currently, manual interpretation is basically adopted for image and video monitoring. Although image and video monitoring can be used to timely understand the condition of the floating objects in the reservoir site, the potential effectiveness is not exerted far enough, and is mainly realized in the following aspects: the manual interpretation workload is large. Because of the large number of reservoirs and the large number of image and video monitoring sites, the workload of manually judging floats one by one is too great for workers of a county level, a city level and even a provincial level reservoir management unit. Therefore, most of image and video information is only taken as a necessary auxiliary means, and important attention is paid when a reservoir is in a problem; the manual interpretation is not in time. Because the number of sites is large, the problem of large-area floaters is difficult to find in time by manual interpretation, evidence cannot be obtained in time and early warning is carried out, so that the normalized float cleaning work efficiency is low, and the ecological supervision capability of water quality is weak. In the last ten years, the development and application of the deep learning bring great influence to various scientific and technological fields and life fields, and the application start of the deep learning image recognition segmentation in the water conservancy industry is late, and is not mature enough compared with other fields. With the development of the deep learning algorithm and the Internet of things, cloud computing capacity is fully developed, potential of video monitoring is developed, intelligent monitoring of reservoir floaters is achieved, and the method has very important significance for reducing cost and enhancing reservoir safety monitoring management.
Disclosure of Invention
The invention aims to: the intelligent detection inspection level of the water surface floaters of the reservoir is improved, an effective intelligent recognition method of the water surface floaters is provided by utilizing an artificial intelligence emerging technology, the labor cost is reduced, and the standardized intelligent management level of the reservoir is greatly improved.
The technical scheme is as follows: in order to achieve the above purpose, the invention provides an automatic reservoir float identification method based on an improved SegNet model, which comprises the following steps:
s1: acquiring reservoir floater image data under different scenes acquired by aerial photography by using an unmanned aerial vehicle, classifying and screening the image data, and removing invalid images to obtain a reservoir floater image data set for training an improved SegNet model;
s2: preprocessing a reservoir float image data set: the data set formed in the step S1 is subjected to noise reduction processing by utilizing a data noise reduction algorithm, so that the accuracy of reservoir floater image information is improved, and meanwhile, the data set is subjected to data enhancement by utilizing a data enhancement algorithm, and the scale of the training data set is expanded;
s3: labeling the data sets, and dividing the data sets into training sets and verification sets according to the number of the labeled data sets in a ratio of 7:3;
s4: performing model training on the improved SegNet algorithm model by utilizing the training set and the verification set determined in the step S3, and outputting an optimal network structure model and various parameters;
s5: detecting and identifying reservoir floaters: detecting the water surface image of the reservoir acquired by the unmanned aerial vehicle by utilizing an improved SegNet algorithm model based on the optimal network structure model parameters, judging whether reservoir floaters needing to be cleaned exist, and if so, calculating and marking the position information and coverage area of the floaters;
s6: and (5) pushing the detection and identification result of the water reservoir floaters in the step (S5) to relevant responsible persons and management staff to help timely clean the reservoir floaters.
Further, the acquisition of the image data in the step S1 is as follows:
the image data distribution needs to comprise different illumination degrees, such as early, middle, late and the like in the day; the image data distribution also needs to include different weather disturbances such as sunny, rainy, cloudy, fog, wind, etc.; the image data distribution also needs to include the influence of surrounding environment and other objects on the water surface, such as surrounding environment reflection of houses, trees, bridges and the like, weeds, water colors, fishing net fish boxes, fishing boats and the like.
Further, the specific content of the data preprocessing in the step S2 is as follows:
the data noise reduction adopts a Gaussian filtering algorithm; the data enhancement adopts a geometric transformation method, and mainly adopts five transformation methods of turning, rotating, cutting, scaling and translation.
Further, the specific algorithm flow of the gaussian filtering is as follows:
s21: performing zero padding on the image;
s22: generating a Gaussian filter template according to the kernel size and standard deviation size of the set Gaussian filter;
s23: gaussian filtering of the input image: scanning each pixel in the image by using the template generated in the step S22, and replacing the value of the central pixel point of the template by using the weighted average gray value of the pixels in the neighborhood determined by the template;
s24: and outputting the Gaussian filtered image.
Further, the specific formula of the two-dimensional gaussian function used to generate the gaussian filter template in step S22 is as follows:
wherein (x, y) is a point coordinate, sigma is a standard deviation, coordinates of each position of the image are brought into a two-dimensional Gaussian function, and the obtained value is a coefficient corresponding to the (x, y) coordinate in the template.
Further, the step S3 performs semantic annotation on the data set as follows:
and carrying out pixel-level semantic segmentation labeling on the reservoir floater image by using a dataset labeling tool LabelMe, continuously clicking and drawing a series of points at the edge of the floater, wherein the starting point and the end point of the points are the same point, and the points are connected in sequence and are connected end to form a closed polygon.
Further, the specific steps of training the model in step S4 are as follows:
s41: the training set is input into an improved SegNet network model of initialization parameters, and the model is specifically improved as follows: the network structure combines the maximum pooling index and residual connection, shallow layer feature mapping extracted in the encoder stage is input into the nonlinear upsampling stage of the decoder, and the characteristics such as color, texture and boundary of an original image are reserved to the greatest extent through dense feature mapping images generated by deconvolution, so that the identification precision of reservoir floaters is effectively improved;
s42: according to the principle of minimum cross entropy loss of the segmentation result, continuously iterating the network update model parameters by using the verification set until convergence and minimum loss are reached, and setting a cross entropy loss verification function as follows:
introducing a balance factor beta, wherein the value is in the interval of [0,1 ]:
the modified Cross Entropy Loss equation (B-CE) is designed as follows:
CE(p，y)＝-βlog(p t )
where p represents the predicted probability that the sample is in that category and y represents the sample label.
S43: and the model parameters reach the minimum loss, the model training is completed, and the optimal network structure model and various parameters are output.
Further, the reservoir float coverage area calculation method in step S5 is as follows:
s51: the image is converted from BGR to HSV, so that color distinction is facilitated;
s52: extracting a specified color by using a cv2.inRange function;
s53: the contours are found and the areas of the contours are calculated.
Further, the specific steps of pushing information in step S6 are as follows:
s61: calculating coordinates of a center point A of a reservoir floating object area: acquiring the circumscribed rectangle of the region, namely acquiring the maximum and minimum X, Y of the region (namely, the coordinates of four corner points of the circumscribed rectangle: XMax, XMin, YMax, YMin), the center point A coordinate is calculated as follows:
CenterX＝(XMax+XMin)/2
CenterY＝(YMax+YMin)/2
s62: placing the areas responsible for all responsible persons into the same coordinate system, and marking the range responsible for all responsible persons;
s63: judging whether the central point A is in a region which is responsible for a certain responsible person by using a ray method, namely taking the point A as a starting point, taking the x-axis direction as the positive direction of rays, judging the number of intersection points of the straight line and a specific region, if the point is an odd point, the point is in the specific region, and if the point is an even point, the point is not in the specific region;
s64: if the central point A is judged to be in the area which is responsible for a responsible person, the information such as the identified coverage area of the reservoir floaters is pushed to the relevant responsible person.
The beneficial effects are that: compared with the prior art, the invention has the following advantages:
1. the improved algorithm model greatly reduces training parameters aiming at the application scene and improves training efficiency.
2. The improved algorithm introduces a multi-residual error connection strategy, more comprehensively reserves a large amount of detail information contained in the multi-scale image, and reduces information loss caused by reduction and downsampling. Meanwhile, a cross entropy loss function with a balance factor is designed, so that the convergence efficiency of network training is further accelerated, and the imbalance problem of samples is improved.
3. The intelligent monitoring of reservoir floaters is realized, so that the cost is reduced and the reservoir safety monitoring management is enhanced.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a network training flow diagram;
FIG. 3 is a schematic diagram of a ray method for dividing the area to which the reservoir float belongs;
Detailed Description
The present invention is further illustrated in the accompanying drawings and detailed description which are to be understood as being merely illustrative of the invention and not limiting of its scope, and various modifications of the invention, which are equivalent to those skilled in the art upon reading the invention, will fall within the scope of the invention as defined in the appended claims.
The invention provides an automatic reservoir floater identification method based on an improved SegNet model, which is shown in figure 1 and comprises the following steps:
s1: and acquiring reservoir floater image data under different scenes acquired by aerial photography by using an unmanned aerial vehicle, classifying and screening the image data, and removing invalid images to obtain a reservoir floater image data set for training an improved SegNet model. The specific requirements for image data acquisition are as follows:
the image data distribution needs to comprise different illumination degrees, such as early, middle, late and the like in the day; the image data distribution also needs to include different weather disturbances such as sunny, rainy, cloudy, fog, wind, etc.; the image data distribution also needs to include the influence of surrounding environment and other objects on the water surface, such as surrounding environment reflection of houses, trees, bridges and the like, weeds, water colors, fishing net fish boxes, fishing boats and the like.
S2: preprocessing a reservoir float image data set: the data set formed in the step S1 is subjected to noise reduction treatment by utilizing a data noise reduction algorithm, so that the accuracy of the image information of the reservoir floaters is improved; and meanwhile, data enhancement is carried out on the data set by utilizing a data enhancement algorithm, and the scale of the training data set is expanded on the data. The dataset preprocessing uses a specific algorithm:
the data noise reduction adopts a Gaussian filtering algorithm; the data enhancement adopts a geometric transformation method, and mainly adopts five transformation methods of turning, rotating, cutting, scaling and translation. The specific algorithm flow of Gaussian filter is as follows:
s21: performing zero padding on the image;
s22: and generating a Gaussian filter template according to the set kernel size and standard deviation size of the Gaussian filter. The specific formula of the two-dimensional gaussian function used to generate the gaussian filter template is as follows:
wherein (x, y) is a point coordinate, sigma is a standard deviation, coordinates of each position of the image are brought into a two-dimensional Gaussian function, and the obtained value is a coefficient corresponding to the (x, y) coordinate in the template.
S23: gaussian filtering of the input image: scanning each pixel in the image by using the template generated in the step S22, and replacing the value of the central pixel point of the template by using the weighted average gray value of the pixels in the neighborhood determined by the template;
s24: and outputting the Gaussian filtered image.
S3: labeling the data sets, and dividing the data sets into training sets and verification sets according to the number of the labeled data sets in a ratio of 7:3. The specific content of semantic annotation on the data set is as follows:
and carrying out pixel-level semantic segmentation labeling on the reservoir floater image by using a dataset labeling tool LabelMe, continuously clicking and drawing a series of points at the edge of the floater, wherein the starting point and the end point of the points are the same point, and the points are connected in sequence and are connected end to form a closed polygon.
S4: and (3) performing model training on the improved SegNet algorithm model by using the training set and the verification set determined in the step (S3), and outputting an optimal network structure model and various parameters. The specific steps of training the model are shown in fig. 2:
s41: the training set is input into an improved SegNet network model of initialization parameters, and the model is specifically improved as follows: the network structure combines the maximum pooling index and residual connection, shallow layer feature mapping extracted in the encoder stage is input into the nonlinear upsampling stage of the decoder, and the characteristics such as color, texture and boundary of an original image are reserved to the greatest extent through dense feature mapping images generated by deconvolution, so that the identification precision of reservoir floaters is effectively improved;
s42: according to the principle of minimum cross entropy loss of the segmentation result, continuously iterating the network update model parameters by using the verification set until convergence and minimum loss are reached, and setting a cross entropy loss verification function as follows:
introducing a balance factor beta, wherein the value is in the interval of [0,1 ]:
the modified Cross Entropy Loss equation (B-CE) is designed as follows:
CE(p，y)＝-βlog(p t )
where p represents the predicted probability that the sample is in that category and y represents the sample label.
S43: and the model parameters reach the minimum loss, the model training is completed, and the optimal network structure model and various parameters are output.
S5: reservoir floater picture detection and identification and floater coverage area calculation: and detecting the water surface image of the reservoir acquired by the unmanned aerial vehicle by utilizing an improved SegNet algorithm model based on the optimal network structure model parameters, judging whether reservoir floaters needing to be cleaned exist, and if so, calculating and marking the position information and the coverage area of the floaters. When the improved SegNet model is used for reservoir floater image identification, the method specifically comprises the following steps:
(1) Convolving the input reservoir float image to obtain characteristic images of H.W.64 channels, denoted as F 1 ；
(2) Will F 1 Downsampling to obtain H/2*W/2×64, and convolving to obtain H/2*W/2×128, denoted as F 2 ；
(3) Will F 2 Downsampling to obtain H/4*W/4×128, and convolving to obtain H/4*W/4×256, denoted as F 3 ；
(4) Will F 3 Downsampling to obtain H/8*W/8×256, denoted as F 4 ；
(5) Will F 4 Upsampling to obtain H/4*W/4X 256, denoted as F' 3 Then F 'is carried out' 3 Characteristic images of H/4*W/4 x 128, denoted De (F) 3 ′)，F′ 3 The calculation formula of (2) is as follows:
F′ 3 ＝Fuse(PI(F 3 )，F 4 )
wherein PI represents the pooling index and Fuse represents the fusion of feature maps;
(6) De (F) 3 ') up-sampling to obtain H/2*W/2 x 128, denoted as F' 2 Then F 'is carried out' 2 H/2*W/2 x 64, denoted De (F' 2 )，F′ 2 The calculation formula of (2) is as follows:
F′ 2 ＝Fuse(PI(F 2 )，De(F′ 3 ))
(7) De (F' 2 ) Restoring resolution to H×W after upsampling, combining F 1 H.W.128, denoted F, is obtained by cascade operation c The calculation formula is as follows:
F c ＝Conc(Fuse(PI(F 1 )，De(F′ 2 ))，F 1 )
(8) And finally, assigning a value to the category to which each pixel belongs through a softmax function, outputting a corresponding semantic segmentation result, and calculating the reservoir float coverage area according to the output reservoir float semantic segmentation result. The calculation method of the reservoir float coverage area comprises the following steps:
(1) The image is converted from RGB to HSV, so that color distinction is facilitated;
(2) Extracting a specified color by using a cv2.inRange function;
(3) The contours are found and the areas of the contours are calculated.
S6: and (5) pushing the detection and identification result of the water reservoir floaters in the step (S5) to relevant responsible persons and management staff to help timely clean the reservoir floaters. The specific steps of pushing information are shown in fig. 3:
s61: calculating coordinates of a center point A of a reservoir float area according to reservoir float position information: acquiring the circumscribed rectangle of the region, namely acquiring the maximum and minimum X, Y of the region (namely, the coordinates of four corner points of the circumscribed rectangle: XMax, XMin, YMax, YMin), the center point A coordinate is calculated as follows:
CenterX＝(XMax+XMin)/2
CenterY＝(YMax+YMin)/2
s62: placing the areas responsible for all responsible persons into the same coordinate system, and marking the range responsible for all responsible persons;
s63: judging whether the central point A is in a region which is responsible for a certain responsible person by using a ray method, namely taking the point A as a starting point, taking the x-axis direction as the positive direction of rays, judging the number of intersection points of the straight line and a specific region, if the point is an odd point, the point is in the specific region, and if the point is an even point, the point is not in the specific region;
s64: if the central point A is judged to be in the area which is responsible for a responsible person, the information such as the identified coverage area of the reservoir floaters is pushed to the relevant responsible person.
Claims (9)
1. The reservoir floater automatic identification method based on the improved SegNet model is characterized by comprising the following steps of:
s1: acquiring reservoir floater image data under different scenes acquired by aerial photography by using an unmanned aerial vehicle, classifying and screening the image data, and removing invalid images to obtain a reservoir floater image data set for training an improved SegNet model;
s2: reservoir floater picture detection and identification and floater coverage area calculation: the data set formed in the step S1 is subjected to noise reduction treatment by utilizing a data noise reduction algorithm, so that the accuracy of the image information of the reservoir floaters is improved; meanwhile, the data enhancement algorithm is utilized to carry out data enhancement on the data set, and the scale of the training data set is expanded on the data;
s3: labeling the data sets, and dividing the data sets into training sets and verification sets according to the number of the labeled data sets in a ratio of 7:3;
s4: performing model training on the improved SegNet algorithm model by utilizing the training set and the verification set determined in the step S3, and outputting an optimal network structure model and various parameters;
s5: detecting and identifying reservoir floaters: detecting the water surface image of the reservoir acquired by the unmanned aerial vehicle by utilizing an improved SegNet algorithm model based on the optimal network structure model parameters, judging whether reservoir floaters needing to be cleaned exist, and if so, calculating and marking the position information and coverage area of the floaters;
s6: and (5) pushing the detection and identification result of the water reservoir floaters in the step (S5) to relevant responsible persons and management staff to help timely clean the reservoir floaters.
2. The reservoir float identification method based on the improved SegNet model of claim 1, wherein the image data acquisition in step S1 is as follows:
the image data distribution needs to comprise different illumination degrees, such as early, middle, late and the like in the day; the image data distribution also needs to include different weather disturbances such as sunny, rainy, cloudy, fog, wind, etc.; the image data distribution also needs to include the influence of surrounding environment and other objects on the water surface, such as surrounding environment reflection of houses, trees, bridges and the like, weeds, water colors, fishing net fish boxes, fishing boats and the like.
3. The reservoir float identification method based on the improved SegNet model according to claim 1, wherein the specific content of the data preprocessing in the step S2 is as follows:
the data noise reduction adopts a Gaussian filtering algorithm; the data enhancement adopts a geometric transformation method, and mainly adopts five transformation methods of turning, rotating, cutting, scaling and translation.
4. The reservoir float automatic identification method based on the improved SegNet model according to claim 3, wherein the specific algorithm flow of Gaussian filtering is as follows:
s21: performing zero padding on the image;
s22: generating a Gaussian filter template according to the kernel size and standard deviation size of the set Gaussian filter;
s23: gaussian filtering of the input image: scanning each pixel in the image by using the template generated in the step S22, and replacing the value of the central pixel point of the template by using the weighted average gray value of the pixels in the neighborhood determined by the template;
s24: and outputting the Gaussian filtered image.
5. The method for automatically identifying reservoir floats based on an improved SegNet model according to claim 4, wherein the specific formula of the two-dimensional Gaussian function used for generating the Gaussian filter template in the step S22 is as follows:
wherein (x, y) is a point coordinate, sigma is a standard deviation, coordinates of each position of the image are brought into a two-dimensional Gaussian function, and the obtained value is a coefficient corresponding to the (x, y) coordinate in the template.
6. The automatic reservoir float identification method based on the improved SegNet model according to claim 1, wherein the semantic annotation of the data set in step S3 is as follows:
and carrying out pixel-level semantic segmentation labeling on the reservoir floater image by using a dataset labeling tool LabelMe, continuously clicking and drawing a series of points at the edge of the floater, wherein the starting point and the end point of the points are the same point, and the points are connected in sequence and are connected end to form a closed polygon.
7. The method for automatically identifying reservoir floaters based on the improved SegNet model as claimed in claim 1, wherein the training model in step S4 specifically comprises the following steps:
s41: the training set is input into an improved SegNet network model of initialization parameters, and the model is specifically improved as follows: the network structure combines the maximum pooling index and residual connection, shallow layer feature mapping extracted in the encoder stage is input into the nonlinear upsampling stage of the decoder, and the characteristics such as color, texture and boundary of an original image are reserved to the greatest extent through dense feature mapping images generated by deconvolution, so that the identification precision of reservoir floaters is effectively improved;
s42: according to the principle of minimum cross entropy loss of the segmentation result, continuously iterating the network update model parameters by using the verification set until convergence and minimum loss are reached, and setting a cross entropy loss verification function as follows:
introducing a balance factor beta, wherein the value is in the interval of [0,1 ]:
the modified Cross Entropy Loss equation (B-CE) is designed as follows:
CE(p，y)＝-βlog(p t )
where p represents the predicted probability that the sample is in that category and y represents the sample label.
S43: and the model parameters reach the minimum loss, the model training is completed, and the optimal network structure model and various parameters are output.
8. The automatic reservoir float identification method based on the improved SegNet model as claimed in claim 1, wherein the reservoir float coverage area calculation method in step S5 is as follows:
s51: the image is converted from RGB to HSV, so that color distinction is facilitated;
s52: extracting a specified color by using a cv2.inRange function;
s53: the contours are found and the areas of the contours are calculated.
9. The automatic reservoir floater identification method based on the improved SegNet model according to claim 1, wherein the specific steps of pushing information in step S6 are as follows:
s61: calculating coordinates of a center point A of a reservoir float area according to reservoir float position information: acquiring the circumscribed rectangle of the region, namely acquiring the maximum and minimum X, Y of the region (namely, the coordinates of four corner points of the circumscribed rectangle: XMax, XMin, YMax, YMin), the center point A coordinate is calculated as follows:
CenterX＝(XMax+XMin)/2
CenterY＝(YMax+YMin)/2
s62: placing the areas responsible for all responsible persons into the same coordinate system, and marking the range responsible for all responsible persons;
s63: judging whether the central point A is in a region which is responsible for a certain responsible person by using a ray method, namely taking the point A as a starting point, taking the x-axis direction as the positive direction of rays, judging the number of intersection points of the straight line and a specific region, if the point is an odd point, the point is in the specific region, and if the point is an even point, the point is not in the specific region;
s64: if the central point A is judged to be in the area which is responsible for a responsible person, the information such as the identified coverage area of the reservoir floaters is pushed to the relevant responsible person.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310069796.5A CN117372942A (en) | 2023-02-03 | 2023-02-03 | Reservoir floater automatic identification method based on improved SegNet model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310069796.5A CN117372942A (en) | 2023-02-03 | 2023-02-03 | Reservoir floater automatic identification method based on improved SegNet model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117372942A true CN117372942A (en) | 2024-01-09 |
Family
ID=89397173
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310069796.5A Pending CN117372942A (en) | 2023-02-03 | 2023-02-03 | Reservoir floater automatic identification method based on improved SegNet model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117372942A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139427A (en) * | 2021-03-12 | 2021-07-20 | 浙江智慧视频安防创新中心有限公司 | Steam pipe network intelligent monitoring method, system and equipment based on deep learning |
| CN113469097A (en) * | 2021-07-13 | 2021-10-01 | 大连理工大学人工智能大连研究院 | SSD (solid State disk) network-based real-time detection method for water surface floating object multiple cameras |
-
2023
- 2023-02-03 CN CN202310069796.5A patent/CN117372942A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139427A (en) * | 2021-03-12 | 2021-07-20 | 浙江智慧视频安防创新中心有限公司 | Steam pipe network intelligent monitoring method, system and equipment based on deep learning |
| CN113469097A (en) * | 2021-07-13 | 2021-10-01 | 大连理工大学人工智能大连研究院 | SSD (solid State disk) network-based real-time detection method for water surface floating object multiple cameras |
Non-Patent Citations (4)
| Title |
|---|
| 小鱼爱吃草灬灬: "物体检测之损失函数：标准交叉熵损失、平衡交叉熵损失、 Focal Loss", pages 1 - 4, Retrieved from the Internet <URL:https://blog.csdn.net/qq_42013574/article/details/105864230> * |
| 李静;王文成;: "基于网格中心点的点在多边形内的高效判定", 软件学报, no. 09, 15 September 2012 (2012-09-15), pages 2481 - 2488 * |
| 杨雪松;蔡碧野;张建明;李峰;: "基于HSV颜色特征和轮廓面积的救生衣检测算法", 计算机工程与应用, no. 03, 31 December 2016 (2016-12-31), pages 184 - 188 * |
| 薛明 等: "基于残差SegNet模型的高分辨率遥感影像分类方法", 桂林理工大学学报, vol. 42, no. 4, 31 December 2022 (2022-12-31), pages 918 - 925 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059758B (en) | Remote sensing image culture pond detection method based on semantic segmentation | |
| CN113177183B (en) | Seawater pollution monitoring and early warning method and system based on ocean remote sensing image | |
| CN110796009A (en) | Method and system for detecting marine vessel based on multi-scale convolution neural network model | |
| CN110751075A (en) | Remote sensing image culture pond detection method based on example segmentation | |
| Lin et al. | Bubble size estimation for flotation processes | |
| CN110060508B (en) | Automatic ship detection method for inland river bridge area | |
| CN108564588B (en) | Built-up area automatic extraction method based on depth features and graph segmentation method | |
| CN110472628B (en) | Improved Faster R-CNN network floater detection method based on video characteristics | |
| CN112365468A (en) | AA-gate-Unet-based offshore wind power tower coating defect detection method | |
| CN111028255A (en) | Farmland area pre-screening method and device based on prior information and deep learning | |
| CN110084302B (en) | Crack detection method based on remote sensing image | |
| CN109886159B (en) | Face detection method under non-limited condition | |
| CN114596500A (en) | Remote sensing image semantic segmentation method based on channel-space attention and DeeplabV3plus | |
| CN112418108A (en) | Remote sensing image multi-class target detection method based on sample reweighing | |
| CN111461006A (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
| CN115452759A (en) | River and lake health index evaluation method and system based on satellite remote sensing data | |
| Zheng et al. | Building recognition of UAV remote sensing images by deep learning | |
| CN113420759B (en) | Anti-occlusion and multi-scale dead fish identification system and method based on deep learning | |
| CN114266947A (en) | Classification method and device based on fusion of laser point cloud and visible light image | |
| CN110097549A (en) | Based on morphologic land, water and air boundary line detecting method, system, medium and equipment | |
| CN102542590B (en) | High-resolution SAR (Synthetic Aperture Radar) image marking method based on supervised topic model | |
| CN117372942A (en) | Reservoir floater automatic identification method based on improved SegNet model | |
| CN113192018B (en) | Water-cooled wall surface defect video identification method based on fast segmentation convolutional neural network | |
| CN113379603B (en) | Ship target detection method based on deep learning | |
| CN115063437A (en) | Mangrove canopy visible light image index characteristic analysis method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |