A dam surface animal intelligent identification and tracking method

Abstract

The application provides a dam surface animal intelligent identification and tracking method, a large number of thermal imaging pictures of animals possibly intruding into a dam range in different postures are collected in advance by using an infrared thermal imaging camera, a deep learning algorithm is used to train thermal imaging samples labeled with animal positions, and automatic detection of animal regions in the thermal imaging pictures is realized; then, intelligent classification of animal body types is realized based on a color step gradient algorithm of the infrared thermal imaging pictures; for large animals, an animal type identification algorithm is constructed by taking body posture characteristics and body surface temperature distribution as characteristic parameters; for small animals, a deep learning model is constructed by taking walking track characteristics and walking speed as characteristic parameters, and intelligent identification of target animal types is realized. The application is highly consistent with the living habits of wild animals, can accurately identify animal types, action tracks and cave positions on the dam at night, and reduces the possibility of dam collapse caused by cave leakage channels.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of dam safety monitoring, and particularly relates to a dam surface animal intelligent identification and tracking method. BACKGROUND

[0002] China is a water conservancy power with vast territory, numerous rivers, and great height difference and river drop, and rich water energy resources. Water conservancy facilities play an increasingly important role in ensuring social and economic stability and people's living standards. Among them, the earth-rock dam has the advantages of local material, simple structure and low cost, and has been developed on a large scale in China, and various earth-rock dams account for more than 95% of the total number of dams in China.

[0003] Unlike concrete gravity dams and concrete arch dams, the surface of earth-rock dams is generally planted with grass to protect the dam surface from soil and stone loss, and has a good ecological environment. However, the excellent environment also attracts various animals to live here, and the animals found in the earth-rock dam area include ants, rodents, snakes, badgers, rabbits, cats, dogs, etc., among which ants, rodents, snakes, badgers and rabbits are used to digging holes in the earth-rock dam to live, and snakes and dogs are harmful to humans. A thousand li of embankment is destroyed by an anthill. The holes on the earth-rock dam may cause a penetrating leakage channel, seriously threatening the stability and safety of the dam, and have the possibility of causing dam failure. The consequences of dam failure are disastrous, causing immeasurable loss to downstream town construction, people's life and property. The Teton Dam in the United States failed during the impoundment process, causing 11 deaths, more than 20,000 people to be displaced, and a large number of farmland and transportation facilities to be destroyed; the Malpasset Dam in France suddenly failed in heavy rain, causing 423 deaths, the downstream Frejus city becoming a ruin, and nearby buildings, roads and power supply facilities being almost completely destroyed.

[0004] The animal holes on the earth-rock dam are generally hidden, and the animals often move and forage at night, making it difficult for dam managers to find them, so it is difficult to check the holes on the earth-rock dam and drive away the animals. The infrared thermal imaging camera has high sensitivity to the temperature of living beings at night, and can identify the animal species on the earth-rock dam according to the temperature distribution or walking track characteristics of the animal body surface, and find the hole position of the animal, solving the problem of animal threat to the safety of the earth-rock dam. SUMMARY

[0005] The first object of the present application is to solve the problem that animals are easy to move and dig holes on the surface of the earth-rock dam, threatening human and dam safety and being difficult to identify, by extracting the method of combining infrared thermal imaging detection technology with a deep learning model for intelligent identification of animals on the dam surface.

[0006] To this end, the above object of the present application is achieved by the following technical solution:

[0007] A dam surface animal intelligent identification method, comprising the following steps:

[0008] S110, collect the target animal information that may appear on the dam, and construct a target animal feature parameter library;

[0009] S120, for different ages and sizes of target animals that may intrude into the dam range, use an infrared thermal imaging camera to collect a large number of thermal imaging pictures of the target animals in different postures in a dark environment, forming a target animal thermal imaging picture sample library; at the same time, collect the thermal imaging pictures of humans as negative samples to exclude the interference of human activities;

[0010] S130, construct a dam target animal intrusion detection model based on a two-stage image target detection algorithm;

[0011] S140, based on whether the infrared thermal imaging camera can clearly display the target animal body surface temperature distribution under certain distance conditions, the target animals are divided into large target animals and small target animals, that is, the target animal region in the infrared thermal imaging camera presents multiple colors, which is determined as a large target animal, and the target animal in the infrared thermal imaging camera presents a single color, which is determined as a small target animal, to form a target animal size classification model;

[0012] S150, for large target animals: construct a large target animal species identification model with body posture features and body surface temperature distribution as feature parameters for intelligent identification of target animal species;

[0013] S160, for small target animals: construct a small target animal species identification model with walking track features as feature parameters, and construct a small target animal walking speed parameter library with walking speed as a feature parameter for intelligent identification of target animal species;

[0014] S170, fuse the target animal feature parameter library, the target animal intrusion detection model, the target animal size classification model, the large target animal species identification model, the small target animal species identification model, and the small target animal walking speed parameter library to form a set of dam surface animal intelligent identification system.

[0015] While adopting the above technical solution, the present application can also adopt or combine the following technical solutions:

[0016] As a preferred technical solution of the present application: step S110 specifically comprises: consulting materials, sorting out target animal species that may appear on the dam or dig holes, collecting basic living characteristics of target animals such as body size, body shape characteristics, body surface temperature, walking speed, walking track, activity time and living habits, and constituting a target animal characteristic parameter library.

[0017] The target animals include but are not limited to rodents, snakes, badgers, rabbits, cats, dogs and other animals that may appear on the dam or dig holes. Some of the target animals' holes have great harmfulness to the dam.

[0018] As a preferred technical solution of the present application: step S130 specifically comprises:

[0019] S131, marking the position and size of the target animal in each thermal imaging picture in the sample library in the form of a rectangular frame;

[0020] S132, based on the classic VGG-16 convolutional neural network structure, adding a standard Batch Normalization layer after each convolutional layer to form an improved VGG-16 network structure for target animal thermal imaging picture feature extraction; constructing a Faster R-CNN target detection framework algorithm and embedding the improved VGG-16 network structure therein to form an algorithm for automatically detecting the target region of the thermal imaging picture;

[0021] S133, dividing the labeled thermal imaging picture samples into training samples and test samples in a ratio of 7:3, training the training samples by using the Faster RCNN target detection framework combined with the improved VGG-16 network structure, and obtaining an automatic detection model of the target animal region in the thermal imaging picture;

[0022] S134, inputting the target animal thermal imaging picture test sample into the detection model to verify the accuracy of the target animal detection.

[0023] As a preferred technical solution of the present application: step S140 specifically comprises:

[0024] S141, for the thermal imaging pictures collected by the infrared thermal imaging camera on the dam, the target animal intrusion detection model in step S130 is used to detect the target animal region in the picture and cut it into an independent thermal imaging picture;

[0025] S142, extracting the three channel (RGB) color step values (0-255) of each pixel point in the cut thermal imaging picture, and calculating the average value of the three channel color step values of each pixel point, the formula is as follows:

[0026]

[0027] wherein R i,j , G i,j and B i,j are the color scale values of the red channel, green channel and blue channel respectively of the pixel at the i-th row and j-th column in the picture, is the average value of the three-channel color scale values of the pixel at the i-th row and j-th column in the picture;

[0028] S143, for each pixel X(i,j) on the thermal imaging picture, the gradient values of the average color scale values of the left neighboring pixel X(i-1,j) and the upper neighboring pixel X(i,j-1) are calculated respectively, and the formula is as follows:

[0029]

[0030]

[0031] wherein, are the average color scale values of the left neighboring pixel and the upper neighboring pixel respectively, d1 and d2 are the pixel spacing in the horizontal direction and the vertical direction respectively, S i,j , C i,j are the average color scale gradient values of the pixel X(i,j) in the horizontal direction and the vertical direction respectively;

[0032] S144, according to the color scale distribution of the animal body surface and the color scale difference between the animal and the background in the target animal thermal imaging picture, a color scale gradient threshold is set;

[0033] S145, for all the pixels X(i,j) in the thermal image, if the average color scale gradient value S i,j in the horizontal direction or the average color scale gradient value C i,j in the vertical direction exceeds the color scale gradient threshold, the pixel is marked;

[0034] S146, for the target animal thermal imaging picture after cropping:

[0035] (1) if the marked pixels only form one closed figure, it is considered that the target animal presents a single color in the infrared thermal imaging camera, and it is determined as a small target animal;

[0036] (2) if the marked pixels form two or more closed figures, it is considered that the target animal presents multiple colors in the infrared thermal imaging camera, and it is determined as a large target animal.

[0037] As a preferred technical solution of the present application: the setting standard of the color gradient threshold is whether the obvious color difference can be seen in the infrared thermal imaging camera under the specific distance condition, and the purpose is to distinguish the large target animals showing multiple colors from the small target animals showing single color.

[0038] As a preferred technical solution of the present application: step S150 specifically comprises:

[0039] S151, selecting all the thermal imaging pictures containing large target animals from the target animal thermal imaging picture sample library obtained in step S120, and cutting out the large target animal region in the thermal imaging picture to form a large target animal thermal imaging picture sample library;

[0040] S152, marking the species of large target animals in each thermal imaging picture in the sample library;

[0041] S153, uniformly dividing the marked large target animal thermal imaging picture sample into training samples and test samples in a ratio of 7:3, training the training samples by using the improved VGG-16 neural network structure in step S132, and obtaining a large target animal species identification model in the thermal imaging picture;

[0042] S154, inputting the large target animal thermal imaging picture test sample into the species identification model to verify the accuracy of the species identification of the large target animals.

[0043] As a preferred technical solution of the present application: step S160 specifically comprises:

[0044] S161, for small target animals with different living habits, using an infrared thermal imaging camera to record a large amount of target animal activity videos for a long time in a dark environment, using a motion tracking algorithm in the OpenCV library to draw the activity track of the target animals in the thermal imaging picture in each video, recording the time spent to complete the walking track, and calculating the length and walking speed of the walking track according to the shooting ratio of the camera, forming a thermal imaging picture sample library of the walking track of the target animals and a walking speed parameter library;

[0045] S162, marking the thermal imaging pictures of the walking track of the target animals in the sample library with the species of the small target animals as the label, and uniformly dividing the marked thermal imaging picture sample into training samples and test samples in a ratio of 7:3;

[0046] S163, training the training samples by using the improved VGG-16 network structure in step S132 to obtain a small target animal species identification model characterized by the activity track in the thermal imaging picture;

[0047] S164, input the thermal imaging picture test sample of the walking track of the small target animal into the category identification model to verify the accuracy of the small target animal category identification.

[0048] The small target animal appears in a single color in the infrared thermal imaging camera, and it is difficult to distinguish the category, therefore, the walking track features of each animal are used for species identification.

[0049] The present application also aims to provide a dam surface animal intelligent tracking method based on the dam surface animal intelligent identification method.

[0050] To this end, the above-mentioned purposes of the present application are achieved by the following technical solutions:

[0051] A dam surface animal intelligent tracking method comprises the following steps:

[0052] S210, a certain number of infrared thermal imaging cameras are arranged on the downstream side of the dam, so as to ensure that the shooting range covers the entire downstream slope of the dam while ensuring the shooting accuracy;

[0053] S220, the infrared thermal imaging camera monitoring video is transmitted back to the host end in real time, and the thermal imaging picture is intercepted at fixed time intervals and input into the dam surface animal intelligent identification system:

[0054] The target animal intrusion detection model is used to determine whether a target animal intrudes into the dam area; if a target animal is detected to intrude, the target animal size classification model is used to determine the size of the intruding animal; if it is a large target animal, the large target animal category identification model is used to determine the species of the animal; if it is a small target animal, the monitoring video of the infrared thermal imaging camera in the corresponding period is intercepted, the motion tracking algorithm in the OpenCV library is used to obtain the activity track of the target animal in the thermal imaging picture, and the thermal imaging picture containing the activity track is input into the small target animal category identification model to determine the species of the intruding animal, and then the small target animal walking speed parameter library is used for verification;

[0055] S230, after the species of the intruding animal is determined, the living habits of the animal are automatically output from the target animal feature parameter library to determine the damage degree of the intruding target animal to the dam or human beings;

[0056] S240, the walking track of the intruding target animal is tracked, if the track suddenly disappears at a certain position on the dam, the disappearing position is the cave position of the target animal, and corresponding measures can be taken for treatment.

[0057] While the above technical solutions are used, the present application can also use or combine the following technical solutions:

[0058] As a preferred technical solution of the present application: in step S230, the living habits of the animals include whether to dig holes and the harm to humans.

[0059] The present application provides a dam surface animal intelligent identification and tracking method, which has the following beneficial effects: a large number of thermal imaging pictures of animals in different postures that may intrude into the dam range are collected in advance using an infrared thermal imaging camera, a deep learning algorithm is used to train the thermal imaging samples labeled with animal positions, and the automatic detection of the animal region in the thermal imaging picture is realized; then, based on whether the infrared thermal imaging camera can clearly display the target animal body surface temperature distribution under certain distance conditions, the intelligent classification of the animal body type is realized based on the color step gradient algorithm of the infrared thermal imaging picture; for large animals, an animal species identification algorithm is constructed with body posture characteristics and body surface temperature distribution as characteristic parameters; for small animals, a deep learning model is constructed with walking track features and walking speed as characteristic parameters, and the intelligent identification of the target animal species is realized. Finally, the harm degree of the target animal to humans and the dam is judged according to the living habits and track features of the target animal, and the animal's cave position is found out. The present application is highly consistent with the living habits of wild animals, can accurately identify the animal species, action track and cave position on the dam at night, and avoid the possibility of dam collapse caused by cave leakage. In addition, the present application is convenient to popularize and apply, and a set of system is basically suitable for animal intelligent identification and automatic tracking of all dams. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 The flow chart of the dam surface animal intelligent identification method and tracking method provided by the present application is shown.

[0061] Figure 2 The improved VGG-16 neural network structure.

[0062] Figure 3 The thermal imaging color step gradient marking schematic diagram of a large target animal with a variety of colors on the body surface.

[0063] Figure 4 The thermal imaging color step gradient marking schematic diagram of a small target animal with a single color on the body surface. DETAILED DESCRIPTION

[0064] The present application is described in further detail with reference to the drawings and specific embodiments.

[0065] A dam surface animal intelligent identification method, comprising the following steps:

[0066] S110, collect the information of target animals that may appear on the dam, and construct a target animal characteristic parameter library;

[0067] Step S110 specifically includes: consulting materials, sorting out target animal species that may appear on the dam or dig holes, collecting basic living characteristics such as target animal size, body shape, body temperature, walking speed, walking trajectory, activity time and living habits, and constituting a target animal characteristic parameter library.

[0068] The target animals include but are not limited to rodents, snakes, badgers, rabbits, cats, dogs and other animals that may appear on the dam or dig holes. Some of the target animals' holes have great harmfulness to the dam.

[0069] S120, for target animals of different ages and sizes that may intrude into the dam range, an infrared thermal imaging camera is used to collect a large number of thermal imaging pictures of the target animals in different postures in a dark environment, forming a target animal thermal imaging picture sample library; at the same time, the thermal imaging pictures of humans are collected as negative samples to exclude the interference of human activities;

[0070] S130, a dam target animal intrusion detection model is constructed based on a two-stage image target detection algorithm;

[0071] Step S130 specifically includes:

[0072] S131, the position and size of the target animals in each thermal imaging picture in the sample library are marked in the form of a rectangular box;

[0073] S132, based on the classic VGG-16 convolutional neural network structure, a standard Batch Normalization layer is added after each convolutional layer to form an improved VGG-16 network structure for target animal thermal imaging picture feature extraction; a Faster R-CNN target detection framework algorithm is constructed, and the improved VGG-16 network structure is embedded therein to form an algorithm for automatically detecting the target area of the thermal imaging picture;

[0074] S133, the labeled thermal imaging picture samples are evenly divided into training samples and test samples in a ratio of 7:3, the Faster RCNN target detection framework combined with the improved VGG-16 network structure is used to train the training samples, and an automatic detection model of the target animal area in the thermal imaging picture is obtained;

[0075] S134, the target animal thermal imaging picture test sample is input into the detection model to verify the accuracy of the target animal detection.

[0076] S140, according to whether the infrared thermal imaging camera can clearly display the temperature distribution of the target animal body surface under the specific distance condition, the target animal is divided into large target animals and small target animals, that is, the target animal region in the infrared thermal imaging camera presents multiple colors, which is determined as a large target animal, and the target animal in the infrared thermal imaging camera presents a single color, which is determined as a small target animal, so as to form a target animal size classification model;

[0077] Step S140 specifically includes:

[0078] S141, for the thermal imaging picture collected by the infrared thermal imaging camera on the dam, the target animal region in the picture is detected by using the target animal intrusion detection model on the dam in step S130, and is cut into an independent thermal imaging picture;

[0079] S142, the three channel (RGB) color step values (0-255) of each pixel point in the cut thermal imaging picture are extracted, and the average value of the three channel color step values of each pixel point is calculated, and the formula is as follows:

[0080]

[0081] Wherein, R i,j , G i,j and B i,j are the color step values of the red channel, the green channel and the blue channel of the pixel point at the i-th row and the j-th column position in the picture, is the average value of the three channel color step values of the pixel point at the i-th row and the j-th column position in the picture;

[0082] S143, for each pixel point X(i,j) on the thermal imaging picture, the gradient values of the average color step values of the left adjacent pixel point X(i-1,j) and the upper adjacent pixel point X(i,j-1) are calculated respectively, and the formula is as follows:

[0083]

[0084]

[0085] Wherein, are the average color step values of the left adjacent pixel point and the upper adjacent pixel point respectively, d1 and d2 are the pixel point spacing in the horizontal direction and the vertical direction respectively, S i,j , C i,j are the average color step gradient values of the pixel point X(i,j) in the horizontal direction and the vertical direction respectively;

[0086] S144, according to the color step distribution of the animal body surface in the target animal thermal imaging picture and the color step difference between the animal and the background, the color step gradient threshold is set;

[0087] S145, for all pixel points X(i, j) in the thermal image, if its horizontal average color step gradient value S i,j Or the vertical average color step gradient value C i,j If it exceeds the color step gradient threshold value, the pixel point is marked;

[0088] S146, for the cropped target animal thermal imaging picture:

[0089] (1) If the marked pixel points form only one closed figure, it is considered that the target animal presents a single color in the infrared thermal imaging camera, and is determined as a small target animal;

[0090] (2) If the marked pixel points form two or more closed figures, it is considered that the target animal presents multiple colors in the infrared thermal imaging camera, and is determined as a large target animal.

[0091] The setting standard of the color step gradient threshold value is whether the obvious color difference in the infrared thermal imaging camera can be seen under certain distance conditions, and the purpose is to distinguish the large target animal presenting multiple colors from the small target animal presenting a single color.

[0092] S150, for large target animals: a large target animal species identification model is constructed with body posture characteristics and body surface temperature distribution as characteristic parameters for intelligent identification of target animal species;

[0093] Step S150 specifically includes:

[0094] S151, select all thermal imaging pictures containing large target animals from the target animal thermal imaging picture sample library obtained in step S120, and cut out the large target animal region in the thermal imaging picture to form a large target animal thermal imaging picture sample library;

[0095] S152, mark the species of the large target animal in each thermal imaging picture in the sample library;

[0096] S153, divide the marked large target animal thermal imaging picture sample into training samples and test samples in a ratio of 7:3, train the training samples using the improved VGG-16 neural network structure in step S132, and obtain a large target animal species identification model in the thermal imaging picture;

[0097] S154, input the large target animal thermal imaging picture test sample into the species identification model to verify the accuracy of the species identification of the large target animal.

[0098] S160, for small target animals: a small target animal species identification model is constructed with walking track features as characteristic parameters, and a small target animal walking speed parameter library is constructed with walking speed as a characteristic parameter for intelligent identification of target animal species;

[0099] Step S160 specifically includes:

[0100] S161, for small target animals with different living habits, a large number of target animal activity videos are recorded for a long time in a dark environment using an infrared thermal imaging camera, a motion tracking algorithm in the OpenCV library is used to draw the activity track of the target animal in the thermal imaging picture in each video, the time spent to complete the walking track is recorded, and the length and walking speed of the walking track are calculated according to the shooting ratio of the camera, forming a thermal imaging picture sample library of the walking track of the target animal and a walking speed parameter library;

[0101] S162, the thermal imaging pictures of the walking track of the target animal in the sample library are labeled with the species of the small target animal as the label, and the labeled thermal imaging picture samples are evenly divided into training samples and test samples in a ratio of 7:3;

[0102] S163, the training samples are trained using the improved VGG-16 network structure in step S132 to obtain a small target animal species identification model with movement track features in the thermal imaging picture as the characteristic;

[0103] S164, the thermal imaging picture test sample of the walking track of the small target animal is input into the species identification model to verify the accuracy of the small target animal species identification.

[0104] The small target animals appear in a single color in the infrared thermal imaging camera, and it is difficult to distinguish the species, therefore, the walking track features of each animal are used for species identification.

[0105] S170, the target animal characteristic parameter library, the target animal intrusion detection model, the target animal size classification model, the large target animal species identification model, the small target animal species identification model, and the small target animal walking speed parameter library are fused to form a set of dam surface animal intelligent identification system.

[0106] The application also provides a dam surface animal intelligent tracking method based on the dam surface animal intelligent identification method described above, comprising the following steps:

[0107] S210, a certain number of infrared thermal imaging cameras are arranged on the downstream side of the dam, which ensures that the shooting range covers the entire downstream slope of the dam while ensuring shooting accuracy;

[0108] S220, the infrared thermal imaging camera monitoring video is transmitted back to the host terminal in real time, and the thermal imaging pictures are intercepted at fixed time intervals and input to the dam surface animal intelligent recognition system:

[0109] If the target animal is detected to have intruded into the dam area, the target animal body size classification model is used to determine the size of the intruding animal; if it is a large target animal, the large target animal species identification model is used to determine the species of the animal; if it is a small target animal, the monitoring video of the infrared thermal imaging camera in the corresponding period is intercepted, the motion tracking algorithm in the OpenCV library is used to obtain the activity track of the target animal in the thermal imaging picture, and the thermal imaging picture containing the activity track is input to the small target animal species identification model to determine the species of the intruding animal, and then the small target animal walking speed parameter library is used for verification.

[0110] S230, after determining the species of the intruding animal, the life habit of the animal is automatically output from the target animal feature parameter library, and the degree of harm of the intruding target animal to the dam or human is determined.

[0111] S240, the walking track of the intruding target animal is tracked, and if the track suddenly disappears at a position on the dam, the disappearing position is the cave position of the target animal, and corresponding measures can be taken for treatment.

[0112] In step S230, the life habit of the animal includes whether to dig holes and the harm to human.

[0113] The above specific embodiments are used to explain and illustrate the present application, and are only preferred embodiments of the present application, but not limit the present application, any modification, equivalent replacement, improvement, etc. of the present application within the spirit and protection scope of the claims of the present application, falls within the protection scope of the present application.

Claims

1. A method for intelligent recognition of animals on a dam surface, characterized in that: The dam surface animal intelligent identification method comprises the following steps: S110, collect target animal information appearing on the dam, and construct a target animal characteristic parameter library; S120, for different ages and sizes of target animals that intrude into the dam range, use an infrared thermal imaging camera to collect a large number of thermal imaging pictures of the target animals in different postures in a dark environment, and form a target animal thermal imaging picture sample library; at the same time, collect thermal imaging pictures of humans as negative samples to exclude the interference of human activities; S130, construct a dam target animal intrusion detection model based on a two-stage image target detection algorithm; S140, based on whether the infrared thermal imaging camera can clearly display the target animal body surface temperature distribution under a specific distance condition, the target animals are divided into large target animals and small target animals, that is, the target animal region in the infrared thermal imaging camera presents multiple colors, which is determined as a large target animal, and the target animal in the infrared thermal imaging camera presents a single color, which is determined as a small target animal, so as to form a target animal size classification model; S150, for large target animals: a large target animal species identification model is constructed based on body posture characteristics and body surface temperature distribution as characteristic parameters for intelligent identification of target animal species; S160, for small target animals: a small target animal species identification model is constructed based on walking track characteristics as characteristic parameters, and a small target animal walking speed parameter library is constructed based on walking speed as characteristic parameters for intelligent identification of target animal species; S170, fuse the target animal characteristic parameter library, the target animal intrusion detection model, the target animal size classification model, the large target animal species identification model, the small target animal species identification model, and the small target animal walking speed parameter library to form a dam surface animal intelligent identification system; Step S130 specifically comprises: S131, mark the position and size of the target animal in each thermal imaging picture in the sample library in the form of a rectangular frame; S132, based on the classic VGG-16 convolutional neural network structure, add a standard Batch Normalization layer after each convolutional layer to form an improved VGG-16 network structure for target animal thermal imaging picture feature extraction; construct a Faster R-CNN target detection framework algorithm, and embed the improved VGG-16 network structure therein to form an algorithm for automatically detecting the target region of the thermal imaging picture; S133, divide the labeled thermal imaging picture samples into training samples and test samples in a ratio of 7:3, train the training samples by using the Faster RCNN target detection framework combined with the improved VGG-16 network structure, and obtain an automatic detection model of the target animal region in the thermal imaging picture; S134, input the target animal thermal imaging picture test sample into the detection model to verify the accuracy of target animal detection; Step S140 specifically comprises: S141, for the thermal imaging pictures collected by the infrared thermal imaging camera on the dam, the target animal region in the picture is detected by the target animal intrusion detection model in step S130, and is cropped into an independent thermal imaging picture; S142, the three channel color scale values of each pixel point in the cropped thermal imaging picture are extracted, and the average value of the three channel color scale values of each pixel point is calculated, and the formula is as follows: wherein R i,j , G i,j , and B i,j are the color scale values of the red channel, the green channel, and the blue channel, respectively, of a pixel at the i-th row and the j-th column position in the picture, is the average value of the three-channel color scale values of the pixel at the i-th row and the j-th column position in the picture. S143, for each pixel point X(i,j) on the thermal imaging picture, the gradient value of the average color scale value of the left adjacent pixel point X(i-1,j) and the upper adjacent pixel point X(i,j-1) is calculated, and the formula is as follows: where RGB i-1,j , RGB i,j-1 are average color step values of the left and upper adjacent pixel points, respectively, d1 and d2 are pixel point distances in the horizontal and vertical directions, respectively, S i,j , C i,j are average color step gradient values of the pixel point X(i,j) in the horizontal and vertical directions, respectively. S144, according to the color scale distribution of the animal body surface in the target animal thermal imaging picture and the color scale difference between the animal and the background, the color scale gradient threshold is set; S145、for all pixels X(i, j) in the thermal image, if its horizontal average color step gradient value S i,j or vertical average color step gradient value C i,j exceeds the color step gradient threshold, mark the pixel point; S146, for the cropped target animal thermal imaging picture: (1) if the marked pixel points form only one closed figure, it is considered that the target animal presents single color in the infrared thermal imaging camera, and it is determined as a small target animal; (2) if the marked pixel points form two or more closed figures, it is considered that the target animal presents multiple colors in the infrared thermal imaging camera, and it is determined as a large target animal. 2.The dam surface animal intelligent identification method of claim 1, wherein: Step S110 specifically includes: consulting materials, sorting out the types of target animals appearing on the dam or digging holes, collecting the size, body characteristics, body temperature, walking speed, walking track, activity time and living habits of the target animals, and constituting a target animal characteristic parameter library. 3.The dam surface animal intelligent identification method of claim 1, wherein: The setting standard of the color scale gradient threshold is whether the obvious color difference can be seen in the infrared thermal imaging camera under certain distance conditions. 4.The dam surface animal intelligent identification method of claim 1, wherein: Step S150 specifically includes: S151, all thermal imaging pictures containing large target animals are selected from the target animal thermal imaging picture sample library obtained in step S120, the large target animal region in the thermal imaging picture is intercepted, and a large target animal thermal imaging picture sample library is formed; S152, the types of large target animals in each thermal imaging picture in the sample library are marked; S153, the large target animal thermal imaging picture samples after marking are evenly divided into training samples and test samples in a ratio of 7:3, an improved VGG-16 neural network structure is used to train the training samples, and a large target animal type identification model in the thermal imaging picture is obtained; S154, the large target animal thermal imaging picture test sample is input into the type identification model, and the accuracy of the type identification of the large target animal is verified. 5.The dam surface animal intelligent identification method of claim 1, wherein: Step S160 specifically includes: S161, for small target animals with different living habits, a large amount of activity video of target animals is recorded for a long time in a dark environment using an infrared thermal imaging camera, the activity track of the target animal in the thermal imaging picture in each video is drawn by using a motion tracking algorithm in an OpenCV library, the time spent for completing the walking track is recorded, and the length and walking speed of the walking track are calculated according to the shooting ratio of the camera, and a thermal imaging picture sample library of the walking track of the target animal and a walking speed parameter library are formed; S162, label the thermal imaging pictures of the target animal walking track in the sample library with small target animal species as a label, and evenly divide the labeled thermal imaging picture samples into training samples and test samples in a ratio of 7:3; S163, train the training samples using an improved VGG-16 network structure to obtain a small target animal species recognition model characterized by the movement track in the thermal imaging picture; S164, input the small target animal walking track thermal imaging picture test sample into the species recognition model to verify the accuracy of the small target animal species recognition.

6. A dam surface animal intelligent tracking method, characterized in that: The dam surface animal intelligent tracking method is based on the dam surface animal intelligent recognition method of claim 1, and comprises the following steps: S210, a certain number of infrared thermal imaging cameras are arranged on the downstream side of the dam to ensure that the shooting range covers the entire downstream slope of the dam while ensuring shooting accuracy; S220, the infrared thermal imaging camera monitoring video is transmitted back to the host end in real time, and the thermal imaging picture is input to the dam surface animal intelligent recognition system according to a fixed time interval: The target animal intrusion detection model is used to determine whether a target animal has intruded into the dam area; if a target animal is detected to have intruded, the target animal size classification model is used to determine the size of the intruding animal; if it is a large target animal, the large target animal species recognition model is used to determine the species of the animal; If it is a small target animal, the monitoring video of the infrared thermal imaging camera in the corresponding period is intercepted, a motion tracking algorithm is used to obtain the activity track of the target animal in the thermal imaging picture, and the thermal imaging picture containing the activity track is input to the small target animal species recognition model to determine the species of the intruding animal, and then the small target animal walking speed parameter library is used for verification; S230, after the species of the intruding animal is determined, the life habits of the animal are automatically output from the target animal characteristic parameter library to determine the degree of harm of the intruding target animal to the dam or human beings; S240, the walking track of the intruding target animal is tracked, and if the track suddenly disappears at a certain position on the dam, the disappearing position is the cave position of the target animal, and corresponding measures are taken for management.

7. The dam surface animal intelligent tracking method of claim 6, wherein: In step S230, the life habits of the animal include whether to dig holes and the harmfulness to human beings.

Images