An unmanned aerial vehicle for identifying animals and plants and its working method

Abstract

The present invention provides a drone for identifying animals and plants and its working method, belonging to the technical field of drones. It solves the problems of long time and low efficiency in identifying animals and plants by drones. The technical solution is as follows: It includes the following steps: Step 1, image acquisition; Step 2, preprocessing of picture information; Step 3, image recognition; Step 4, contour recognition; Step 5, data comparison; Step 6, alarm; The drone for identifying animals and plants includes a drone body; an image acquisition structure is arranged below the drone body. The beneficial effects of the present invention are: effectively improving the efficiency and accuracy of animal and plant identification, and combining a thermal imager and a high-definition camera to improve the identification accuracy during the tracking of animals and plants.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and in particular, to an unmanned aerial vehicle for identifying animals and plants and a working method thereof. Background Art

[0002] The survival and reproduction of human beings are inseparable from wetlands. The wetland ecological environment is complex and has a wide distribution area. Each species is an important member of the ecosystem. Through the food chain relationship, species depend on and restrict each other. However, once a problem occurs in a certain link, the balance of the entire ecosystem will be seriously affected. Therefore, in order to ensure biodiversity and improve the ecosystem status, it is crucial to timely and accurately understand the spatial distribution information and living environment status of wetland target animals and plants. At present, the detection methods of wetland animals and plants include manual detection and remote sensing detection. However, due to the complex ecological environment, manual detection is restricted and not easy to carry out, and remote sensing detection is more commonly used. Compared with optical remote sensing satellites, unmanned aerial vehicles have the advantages of high efficiency and low cost. Therefore, choosing to use unmanned aerial vehicles to collect images in wetland areas and perform target detection and tracking is a more efficient solution.

[0003] In recent years, the technology of unmanned aerial vehicles has developed rapidly and has been widely applied. Among them, the target detection of unmanned aerial vehicle aerial images has been applied in multiple fields. It can not only capture intuitive and highly real-time information, has good timeliness, but also has a wide observation range, is not affected by terrain, and at the same time reduces the investment of human and material resources, which is extremely convenient and provides a powerful means for the detection of wetland animals and plants.

[0004] The application of target detection of unmanned aerial vehicle aerial images is rich, but traditional unmanned aerial vehicle image capture mostly uses the method of photography and is easily affected by the environment. If it encounters rainy, foggy weather or low light at night, the collected image information is not easy to be recognized, which has certain limitations; traditional unmanned aerial vehicle detection mostly focuses on manual operation and manual feature recognition, and a large amount of time is required for target extraction and detection. Later, with the rapid development of computers and artificial intelligence, target detection based on deep learning algorithms has more advantages. They can directly predict the position and category of the target, have a faster detection speed, but at the same time, the complex and diverse background information is also easy to confuse the target to be detected, and it is difficult for general target detection algorithms to achieve ideal detection effects. Researchers have conducted a large number of studies.

[0005] Through domestic literature and patent searches, it was found that in the existing patent "An Unmanned Aerial Vehicle Remote Sensing Mapping Device for Pollution Source Monitoring" (Application No.: CN202120660793.5), an unmanned aerial vehicle remote sensing mapping device for pollution source monitoring is disclosed, including a main body and a camera, and also including a mobile adjustment structure for facilitating the adjustment of the light source position. The adjustment structure is installed on both sides of the inner bottom end. Using a lighting lamp and a threaded rod, the motor rotates to drive the threaded rod to rotate. When the threaded rod rotates, it can drive the position of the movable block to change. The change in the movable position can change the lighting end at the bottom to move left and right, so as to better map the part that needs to be illuminated. However, when using the lighting lamp, this method may affect the accuracy of the image, and even affect the spatial distribution of wetland animals and plants. Moreover, when the surrounding environment is similar in color to the target animals and plants to be detected, it is very difficult to map and identify through this method, and the scope of use has limitations and is not suitable for detecting wetland animals and plants.

[0006] Through domestic literature and patent searches, it was found that in the existing patent "An Unmanned Aerial Vehicle Remote Sensing Detection Method for Pine Wood Nematode Disease Based on a Deep Learning Model" (Application No.: CN202210238878.3), an unmanned aerial vehicle remote sensing monitoring method for pine wood nematode disease based on a deep learning model is disclosed. The method includes: Step 1, obtaining and processing unmanned aerial vehicle remote sensing data, and conducting on-site investigations to understand the actual situation of the operation area: obtaining the data required for pine wood nematode disease monitoring; Step 2, establishing a pine wood nematode disease sample library: obtaining the corresponding pine wood nematode disease labels to construct a pine wood nematode disease sample library; Step 3, designing a targeted deep learning monitoring algorithm for pine wood nematode disease: extracting the target spatial detail information through a spatial information retention module, using a context information module to obtain context information, and combining an attention optimization module to fuse multi-level features and output the final extraction result. However, the fusion of multi-level features of this method has certain difficulties for real-world target extraction because the activity range of animals detected in wetlands is large and the target space is difficult to determine, which has limitations.

[0007] How to solve the above technical problems is the subject faced by the present invention. Summary of the Invention

[0008] The purpose of the present invention is to provide an unmanned aerial vehicle for animal and plant identification and its working method. It solves the problem of difficult identification of animals and plants by unmanned aerial vehicles. It effectively improves the efficiency and accuracy of animal and plant identification. It can combine a thermal imager and a high-definition camera to improve the identification accuracy during the tracking of animals and plants, and solves the problems of long time and low efficiency in the existing identification mode.

[0009] The idea of the present invention is as follows: Images are obtained by combining thermal imaging and high-definition imaging, which reduces the influence of the environment on the identification of animals and plants. After the images are collected, image preprocessing is used to group the images, making the spatial resolution of the thermal imaging and the high-definition images similar. According to the different received thermal radiation energies, a thermal imaging temperature curve is drawn to distinguish warm objects from inanimate backgrounds, making the identification of animals and plants more targeted and objective, and effectively improving the efficiency of the identification of animals and plants.

[0010] In order to achieve the above-mentioned invention purpose, the technical solution adopted by the present invention is specifically as follows:

[0011] A working method of an unmanned aerial vehicle for the identification of animals and plants, comprising the following steps:

[0012] Step 1, image acquisition: An image acquisition module is used to collect the captured images of the thermal imaging and the high-definition camera, and the identification area is photographed from different angles;

[0013] Step 2, preprocessing of picture information: The formed picture information is preprocessed: Through an image preprocessing module, which is used to group the images according to the image position attributes, and group them from the similar perspectives of the thermal imaging and the high-definition pictures at the same moment, and process the pictures to make the spatial resolution of the thermal imaging and the high-definition pictures similar, and make the pixel size, pixel resolution and field of view angle of the thermal imaging and the high-definition pictures similar; According to the different received thermal radiation energies, the thermal imaging separates the warm objects from the inanimate background objects and encodes the warm objects; Using the first channel: Automatically compare the encoded warm objects with the high-definition pictures taken by the camera, and according to the encoding, first separate the target objects from the background objects;

[0014] Step 3, image recognition: The images processed in Step 2 are divided into a part with feature extraction and a part without image extraction; If there is a part with feature extraction, then obtain the pattern features and the neural network classification ability to identify the target image, calculate the coincidence degree with the input pictures in the information database, and submit the result;

[0015] Step 4, contour recognition: Compare according to the contour. If the pixel is smaller than the minimum value that can be compared by the pixel and the comparison is difficult, then magnify it in the same proportion;

[0016] Compare the contour coincidence degree. If the contour coincidence degree is greater than 80%, then pack and send the target information, generate an alarm, and perform associated tracking video recording;

[0017] If the contour coincidence degree is less than 80%, put the target object in the thermal imaging into the corresponding scene in the picture;

[0018] If the coincidence degree of the unobscured part is greater than 80%, then pack and send the target information, generate an alarm, and perform associated tracking video recording;

[0019] If the contour coincidence degree continues to be less than 80%, it is possible that the object is completely obscured, then enter the second channel for manual comparison and further photograph the surroundings of the object according to the instruction.

[0020] Step Five: Data comparison; through appearance recognition and behavior recognition, process the video in slow motion, analyze animal behavior, and compare it with the habits in the plant and animal information database.

[0021] Secondly, through the recognition of the surrounding environment, compare the photographed surrounding environment with the habitats of plants and animals in the information database; if the recognition rate reaches 90% at this time, it will be automatically classified, and if it is lower than 90%, it will be uploaded for manual recognition.

[0022] Step Six: Alarm; classify the encoded warm object, and if any omission is found, an alarm will be generated.

[0023] In the above-mentioned Step Three, if there is no feature extraction part, directly convert the entire image into digital information for input and judge whether there is noise.

[0024] If there is noise, use the infrared image recognition method based on fractal features to extract its fractal features for further texture segmentation and target recognition, perform infrared image preprocessing on it to enhance its contrast, extract the fractal features of the infrared image, and perform infrared image recognition based on the neural network. Through the training of the original data, obtain the optimal weight coefficients, and thus obtain the recognition result.

[0025] If there is no noise, use the image recognition method based on wavelet moments, use the BP network for recognition. After normalizing, polarizing, and extracting the rotation-invariant wavelet moment features of the input image, send it to the BP network classifier for recognition, and thus obtain the recognition result; The BP network, whose full name is Back Propagation, is a multi-layer feedforward network trained by the error backpropagation algorithm and is one of the most widely used neural network models at present. The BP network can learn and store a large number of input-output pattern mapping relationships without revealing the mathematical equations describing this mapping relationship in advance.

[0026] If the coincidence degree can be successfully calculated from the recognition result, directly submit the result, and if the recognition cannot be successfully performed, submit it for manual recognition.

[0027] The drone for plant and animal recognition includes a drone body; an image acquisition structure is arranged below the drone body.

[0028] The image acquisition structure includes a rotatable camera support frame and a pan-tilt camera; the rotatable camera support frame is connected below the image acquisition structure, and the pan-tilt camera is connected below the rotatable camera support frame.

[0029] The pan-tilt camera has three-spectrum heavy-duty functions of infrared thermal imaging, high-definition visible light, and laser supplementary light.

[0030] It also includes the image acquisition module, image preprocessing module, image recognition module, coincidence degree screening module, and target processing module installed on the UAV body, where

[0031] The image acquisition module is used to collect the captured pictures of the thermal imaging and high-definition cameras;

[0032] The image preprocessing module is used to group the images according to the image position and make the spatial resolutions of the thermal imaging and high-definition pictures similar;

[0033] The image recognition module is used to compare the image obtained by visual recognition with the original input image in the system to obtain the contour coincidence degree of the recognition object;

[0034] The target processing module is used to compare with the animal information database and process according to the comparison situation.

[0035] The image acquisition module is connected to the image preprocessing module, and the image preprocessing module includes a separated encoding target object module; the image acquisition module is connected to collect the captured pictures of the thermal imaging and high-definition cameras, and can shoot the recognition area at different angles; the image preprocessing module is used to group the images according to the image position attributes, group them from the same moment similar perspective of the thermal imaging and high-definition pictures, and process to make the spatial resolutions of the thermal imaging and high-definition pictures similar, that is, make the pixel sizes, image resolutions, and field of view angles of the thermal imaging and high-definition pictures similar; the separated encoding target object module is used to separate the warm objects and inanimate backgrounds according to the different received thermal radiation energies, and encode the warm objects and inanimate target objects respectively.

[0036] The image preprocessing module is connected to the image recognition module, and the image recognition module includes a feature extraction module and a non-feature extraction module;

[0037] The feature extraction module is connected to the neural network classification module, and the neural network classification module is connected to a coincidence degree calculation module;

[0038] The feature extraction module is connected to a digital information conversion module, and the digital information conversion module is connected to a noise judgment module. The noise judgment module judges the digital information. When there is no noise or the noise is very small and can be ignored, the noise judgment module is connected to an image recognition method based on wavelet moments. The image recognition method based on wavelet moments is connected to a BP network recognition module. The BP network recognition module is connected to a resolution test module, and the resolution test module is connected to an identification module. When there is noise, that is, the noise cannot be ignored, the noise judgment module is connected to an infrared image recognition method based on fractal features. The infrared image recognition method based on fractal features is connected to an infrared image preprocessing module. The infrared image preprocessing module is connected to an infrared image fractal feature extraction module, and the infrared image fractal feature extraction module is connected to an identification module. The identification module is connected to a coincidence degree judgment module, and the coincidence degree judgment module is connected to a result submission module. The coincidence degree judgment module is connected to a manual identification module.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] 1. By setting an image acquisition module and an image preprocessing module, the image acquisition module can collect the captured pictures of the thermal imaging and the high-definition camera, and can shoot the recognition area at different angles. The image preprocessing module groups the images according to the image position attributes, groups them from the similar perspective of the thermal imaging and the high-definition pictures at the same moment, and processes them to make the spatial resolution of the thermal imaging and the high-definition pictures similar, that is, the pixel size, image resolution and field of view angle of the thermal imaging and the high-definition pictures are similar, effectively improving the efficiency of animal and plant recognition.

[0041] 2. By setting a separated coding target object module, the separated coding target object module is used to draw a thermal imaging temperature curve according to the different received thermal radiation energies, separate the warm objects from the inanimate background, and code the warm objects and the inanimate target objects respectively, making the animal and plant recognition more targeted and goal-oriented.

[0042] 3. Through the image recognition module set in the present invention, namely, the nervous system recognition, in the part with feature extraction, the experience of people is fully utilized to obtain pattern features and the neural network classification ability to identify the target image, making the recognition result more accurate. If there is no feature extraction part, in the image recognition method based on wavelet moments, the BP network is used for recognition. After the input image is normalized, polar coordinated, and the rotation-invariant wavelet moment features are extracted, it is sent to the BP network classifier for recognition to obtain the recognition result. Wavelet moment features have good resolution ability for samples with translation, scaling, and rotation. In the case of no noise, wavelet moment features can correctly distinguish test samples, and the recognition rate is better than that of geometric moments, with a gap reaching 30 percentage points. Since wavelet moments have good ability to extract local features of images, the highest correct recognition rate reaches 98%, so the accuracy of the recognition result can be guaranteed. Among them, in the infrared image recognition method based on fractal features, its fractal features are extracted for further texture segmentation and target recognition. Through the training of the original data, the optimal weight coefficients are obtained, and good recognition results are achieved. Through these several recognition methods, the recognition efficiency and recognition result can be effectively improved.

[0043] 4. Through the overlap degree screening module set in the present invention, for the overlap degree screening module, if the contour overlap degree is greater than 80%, the target information is packaged and sent, an alarm is generated, and associated tracking recording is carried out; if the contour overlap degree is less than 80%, it may be that part of the object is covered. The target object of the thermal imaging can be used to fill the corresponding scene of the high-definition picture. If the overlap degree of the uncovered part is greater than 80%, the target information is packaged and sent, and associated recording is carried out; if the contour overlap degree continues to be less than 80%, it may be that the object is completely covered, and manual intervention can be carried out, and the surrounding of the suspected target can be further photographed according to the instruction, making the recognized target object more accurate.

[0044] 5. Through the target processing module in the present invention, the tracked and photographed animals and plants are compared with the animal and plant information database. Through appearance recognition and behavior recognition, that is, the shooting video is processed in slow motion to analyze the animal behavior and compare it with the habits in the animal and plant information database. If the recognition rate reaches more than 90%, it can be automatically classified. An alarm is given for the missing or unprocessed target objects, and real-time positioning and manual intervention are carried out. In this way, manual operation can be reduced, and recognition and classification can be carried out faster and more efficiently through intelligent analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings are used to provide further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention.

[0046] Figure 1 It is a schematic logical structure diagram of a new type of unmanned aerial vehicle for animal and plant recognition according to the present invention;

[0047] Figure 2 Schematic diagram of the structure of a new UAV image preprocessing module for animal and plant identification according to the present invention;

[0048] Figure 3 Schematic diagram of the structure of a new UAV separation and coding target object module for animal and plant identification according to the present invention;

[0049] Figure 4 Schematic diagram of the algorithm structure of a new UAV image recognition for animal and plant identification according to the present invention;

[0050] Figure 5 Logic structure diagram of a new UAV contour screening module for animal and plant identification according to the present invention;

[0051] Figure 6 Structure diagram of a new UAV for animal and plant identification according to the present invention.

[0052] Among them, the reference numerals are: 1 - UAV body, 2 - image acquisition structure, 21 - rotatable camera support frame, 22 - pan-tilt camera. Specific embodiments

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Of course, the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] Embodiment 1

[0055] As Figures 1-5 shown, the technical solution provided by the present invention is: a working method of a UAV for animal and plant identification, including the following steps:

[0056] Step 1, image acquisition. The image acquisition module is used to collect the pictures taken by the thermal imaging and high-definition cameras, and shoot the identification area at different angles;

[0057] Step 2, picture information preprocessing; the formed picture information is preprocessed: through the image preprocessing module, which is used to group the images according to the image position attributes, and group them according to the similar perspectives of the thermal imaging and high-definition pictures at the same moment, and process the pictures to make the spatial resolutions of the thermal imaging and high-definition pictures similar, and make the pixel sizes, image resolutions and field of view angles of the thermal imaging and high-definition pictures similar; the thermal imaging distinguishes the warm objects from the inanimate background objects according to the different received thermal radiation energies, and encodes the warm objects; using the first channel: automatically compares the encoded warm objects with the high-definition pictures taken by the camera, and according to the encoding, first separates the target objects from the background objects;

[0058] Step 3: Image recognition; divide the image processed in Step 2 into a part with feature extraction and a part without image extraction; if there is a part with feature extraction, obtain the pattern features and the neural network classification ability to recognize the target image, calculate the coincidence degree between the recognized result and the input pictures in the information database, and submit the result;

[0059] Step 4: Contour recognition; perform comparison according to the contour. If the pixel is smaller than the minimum value that can be compared by the pixel and the comparison is difficult, magnify it in proportion;

[0060] Compare the contour coincidence degree. If the contour coincidence degree is greater than 80%, pack and send the target information, generate an alarm, and conduct associated tracking video recording;

[0061] If the contour coincidence degree is less than 80%, put the target object in the thermal imaging into the corresponding scene in the picture;

[0062] If the coincidence degree of the unobscured part is greater than 80%, pack and send the target information, generate an alarm, and conduct associated tracking video recording;

[0063] If the contour coincidence degree continues to be less than 80%, it is possible that the object is completely obscured, then enter the second channel for manual comparison and further photograph the surrounding of the object according to the instruction;

[0064] Step 5: Data comparison; through appearance recognition and behavior recognition, process the video in slow motion, analyze the animal behavior, and compare it with the habits in the animal and plant information database;

[0065] Secondly, through the surrounding environment recognition, compare the photographed surrounding environment with the habitats of animals and plants in the information database; if the recognition rate reaches 90% at this time, it will be automatically classified, and if it is lower than 90%, it will be uploaded for manual recognition;

[0066] Step 6: Alarm; classify the encoded warm object. If any omission is found, generate an alarm.

[0067] In the above Step 3, if there is no part with feature extraction, directly convert the entire image into digital information for input and judge whether there is noise;

[0068] If there is noise, use the infrared image recognition method based on fractal features to extract its fractal features for further texture segmentation and target recognition, perform infrared image preprocessing on it to enhance its contrast, extract the fractal features of the infrared image, and perform infrared image recognition based on the neural network. Through the training of the original data, obtain the optimal weight coefficient, and thus obtain the recognition result;

[0069] If there is no noise, the image recognition method based on wavelet moments is used. The BP network is used for recognition. After the input image is normalized, polar coordinate transformed, and the rotation-invariant wavelet moment features are extracted, it is sent to the BP network classifier for recognition to obtain the recognition result. The BP network, whose full name is Back Propagation, is a multi-layer feedforward network trained by the error backpropagation algorithm and is one of the most widely used neural network models at present. The BP network can learn and store a large number of input-output pattern mapping relationships without revealing the mathematical equations describing this mapping relationship in advance.

[0070] If the coincidence degree can be successfully calculated from the recognition result, the result is directly submitted. If the recognition cannot be successfully performed, it is submitted to humans for recognition.

[0071] Combined Figure 6 , the unmanned aerial vehicle for plant and animal recognition includes the unmanned aerial vehicle body 1; an image acquisition structure 2 is arranged below the unmanned aerial vehicle body 1;

[0072] The image acquisition structure 2 includes a rotatable camera support frame 21 and a pan-tilt camera 22; the rotatable camera support frame 21 is connected below the image acquisition structure 2, and the pan-tilt camera 22 is connected below the rotatable camera support frame 21.

[0073] The pan-tilt camera 22 has the functions of infrared thermal imaging, high-definition visible light, and laser supplementary rice three-spectrum heavy type.

[0074] It also includes the image acquisition module, image preprocessing module, image recognition module, coincidence degree screening module, and target processing module installed on the unmanned aerial vehicle body 1, where:

[0075] The image acquisition module is used to collect the captured pictures of the thermal imaging and high-definition cameras;

[0076] The image preprocessing module is used to group the images according to the image positions and make the spatial resolutions of the thermal imaging and high-definition pictures similar;

[0077] The image recognition module is used to compare the images obtained by visual recognition with the original input images in the system to obtain the contour coincidence degree of the recognition object;

[0078] The target processing module is used to compare with the animal information database and process according to the comparison situation.

[0079] The image acquisition module is connected to an image preprocessing module, and the image preprocessing module includes a module for separating and encoding target objects; the image acquisition module is connected to capture pictures taken by a thermal imaging camera and a high-definition camera, and can capture images of the recognition area from different angles; the image preprocessing module is used to group images according to the image position attributes, group them from the similar perspectives of thermal imaging and high-definition pictures at the same time, and process them to make the spatial resolutions of the thermal imaging and high-definition pictures similar, that is, the pixel sizes, image resolutions, and field of view angles of the thermal imaging and high-definition pictures are similar; the module for separating and encoding target objects is used to distinguish warm objects from inanimate backgrounds according to different received thermal radiation energies, and encode warm objects and inanimate target objects respectively.

[0080] The image preprocessing module is connected to an image recognition module, and the image recognition module includes a feature extraction module and a non-feature extraction module;

[0081] The feature extraction module is connected to a neural network classification module, and the neural network classification module is connected to a module for calculating the coincidence degree;

[0082] The non-feature extraction module is connected to a digital information conversion module, the digital information conversion module is connected to a noise judgment module, and the noise judgment module judges the digital information. If there is no noise or the noise is very small and can be ignored, the noise judgment module is connected to an image recognition method based on wavelet moments, the image recognition method based on wavelet moments is connected to a BP network recognition module, the BP network recognition module is connected to a resolution test module, and the resolution test module is connected to an identification module; if there is noise, that is, the noise cannot be ignored, the noise judgment module is connected to an infrared image recognition method based on fractal features, the infrared image recognition method based on fractal features is connected to an infrared image preprocessing module, the infrared image preprocessing module is connected to an infrared image fractal feature extraction module, and the infrared image fractal feature extraction module is connected to an identification module; the identification module is connected to a coincidence degree judgment module, and the coincidence degree judgment module is connected to a result submission module; the coincidence degree judgment module is connected to an artificial identification module.

[0083] Refer to Figure 6 , which is an unmanned aerial vehicle for identifying animals and plants, including an unmanned aerial vehicle body 1; an image acquisition structure 2 is arranged below the unmanned aerial vehicle body 1;

[0084] The pan-tilt camera 22 uses a three-spectrum heavy-duty integrated camera of infrared thermal imaging + high-definition visible light + laser filling.

[0085] Working principle: The drone rotates the camera support frame 21 to rotate the infrared thermal imaging + high-definition visible light + laser-assisted three-spectrum heavy-duty integrated gimbal camera 22 to take real-time pictures of the object to be measured. The image acquisition structure 2 is used to collect the pictures taken by the thermal imaging and high-definition cameras, and take pictures of the recognition area at different angles.

[0086] After the image acquisition of the drone 1 is completed, the images need to be grouped according to the image position attributes, and preprocessed to make the spatial resolutions of the thermal imaging and high-definition pictures similar. According to the different thermal radiation energies received, the thermal imaging can automatically compare the warm objects with the unencoded warm objects and the high-definition pictures taken by the camera. According to the encoding, the target object and the background object are first separated.

[0087] After the encoding of the drone 1 is completed, image recognition is carried out. After the preprocessed images are judged, they are divided into a part with feature extraction and a part without image extraction. If there is a part with feature extraction, the experience of people is fully utilized to obtain the pattern features and the neural network classification ability to identify the target image, and then the coincidence degree is calculated with the input pictures in the information database, and the result is submitted. If there is no part with feature extraction, the entire image is directly converted into digital information for input, and then it is judged whether there is noise. If there is noise, the infrared image recognition method based on fractal features is used to extract its fractal features for further texture segmentation and target recognition. The infrared image is preprocessed to enhance its contrast, and then the fractal features of the infrared image are extracted. For the infrared image recognition based on the neural network, the best weight coefficients are obtained through the training of the original data, and thus the recognition result is obtained. If there is no noise, the image recognition method based on wavelet moments is used, and the BP network is used for recognition. After the input image is normalized, polar coordinate transformation is performed, and the rotation-invariant wavelet moment features are extracted and then sent to the BP network classifier for recognition, so as to obtain the recognition result. If the coincidence degree can be successfully calculated for the recognition result, the result is directly submitted. If the recognition cannot be carried out smoothly, it is submitted for manual recognition.

[0088] According to the comparison of the contours, if the pixel is smaller than the minimum value that can be compared for the pixel, the comparison is difficult, and it is enlarged proportionally. If the contour coincidence degree is greater than 80%, the target information is packaged and sent, an alarm is generated, and associated tracking recording is carried out. If the contour coincidence degree is less than 80%, it is possible that part of the object is covered. The target object in the thermal imaging can be placed in the corresponding scene in the picture. If the coincidence degree of the uncovered part is greater than 80%, the target information is packaged and sent, an alarm is generated, and associated tracking recording is carried out. If the contour coincidence degree continues to be less than 80%, it is possible that the object is completely covered, then it enters the second channel for manual comparison, and further pictures are taken around the object according to the instructions.

[0089] Compare the tracked animals with the animal information database. First, conduct identification through appearance, including the color and shape of animals and plants, as well as behavior identification. Process the video in slow motion to analyze animal behavior and compare it with the habits in the animal and plant information database. Secondly, conduct identification through the surrounding environment, comparing the photographed surrounding environment with the habitats of animals and plants in the information database. At this time, if the recognition rate reaches 90%, it will be automatically classified; if it is lower than 90%, it will be uploaded for manual identification. Classify the encoded warm object objects and trigger an alarm if any omission is found.

[0090] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A working method of an unmanned aerial vehicle for identifying animals and plants, characterized in that, it includes an unmanned aerial vehicle body (1); an image acquisition structure (2) is arranged below the unmanned aerial vehicle body (1); the image acquisition structure (2) includes a rotatable camera support frame (21) and a pan-tilt camera (22); the rotatable camera support frame (21) is connected below the image acquisition structure (2), and the pan-tilt camera (22) is connected below the rotatable camera support frame (21); the working method includes the following steps: Step 1, image acquisition; use the image acquisition module to collect the pictures taken by the thermal imaging and high-definition cameras, and take pictures of the identification area at different angles; Step 2, preprocessing of picture information; preprocess the formed picture information: through the image preprocessing module, group the images according to the image position attributes, group them by the similar perspectives of thermal imaging and high-definition pictures at the same moment, process the pictures to make the spatial resolution of thermal imaging and high-definition pictures similar, and make the pixel size, image resolution and field of view angle of thermal imaging and high-definition pictures similar; according to the different thermal radiation energies received, thermal imaging separates warm objects from inanimate background objects and encodes the warm objects; use the first channel: automatically compare the encoded warm objects with the high-definition pictures taken by the camera, and according to the encoding, first separate the target object from the background object; Step 3, image recognition; divide the image processed in Step 2 into a part with feature extraction and a part without image extraction; if there is a part with feature extraction, then obtain the pattern features and the neural network classification ability to identify the target image, calculate the coincidence degree with the input pictures in the information database, and submit the result; Step 4, contour recognition; compare according to the contour. If the pixel is smaller than the minimum value that can be compared by the pixel, and the comparison is difficult, then enlarge it in the same proportion; Compare the contour coincidence degree. If the contour coincidence degree is greater than 80%, then pack and send the target information, generate an alarm, and perform associated tracking video recording; If the contour coincidence degree is less than 80%, put the target object in the thermal imaging into the corresponding scene in the picture; If the coincidence degree of the unobscured part is greater than 80%, then pack and send the target information, generate an alarm, and perform associated tracking video recording; If the contour coincidence degree continues to be less than 80% and the object is completely obscured, then enter the second channel for manual comparison, and further photograph the surroundings of the object according to the instruction; Step 5, data comparison; through appearance recognition and behavior recognition, perform slow motion processing on the video, analyze animal behavior, and compare it with the habits in the animal and plant information database; Secondly, through surrounding environment recognition, compare the photographed surrounding environment with the habitats of animals and plants in the information database; at this time, if the recognition rate reaches 90%, it will be automatically classified, and if it is lower than 90%, it will be uploaded for manual recognition; Step 6, alarm; classify the encoded warm object objects, and if any omission is found, generate an alarm.

2. The working method of an unmanned aerial vehicle for identifying animals and plants according to claim 1, characterized in that, the pan-tilt camera (22) has the functions of infrared thermal imaging, high-definition visible light and laser supplementary rice three spectra and is heavy-duty.

3. The working method of an unmanned aerial vehicle for identifying animals and plants according to claim 2, characterized in that, it further includes an image acquisition module, an image preprocessing module, an image recognition module, a coincidence degree screening module, and a target processing module installed on the unmanned aerial vehicle body (1), where the image acquisition module is used to acquire the captured pictures of the thermal imaging and high-definition cameras; the image preprocessing module is used to group the images according to the image positions and make the spatial resolutions of the thermal imaging and high-definition pictures similar; the image recognition module is used to compare the image obtained by visual recognition with the images originally entered in the system to obtain the contour coincidence degree of the recognition object; the target processing module is used to compare with the animal information database and process according to the comparison situation.

4. The working method of an unmanned aerial vehicle for identifying animals and plants according to claim 3, characterized in that, the image acquisition module is connected to the image preprocessing module, and the image preprocessing module includes a module for separating and encoding the target object; the image acquisition module is connected to acquire the captured pictures of the thermal imaging and high-definition cameras, and takes pictures of the recognition area from different angles; the image preprocessing module is used to group the images according to the image position attributes, group them from the similar perspectives of the thermal imaging and high-definition pictures at the same time, and process to make the spatial resolutions of the thermal imaging and high-definition pictures similar, and the pixel sizes, image resolutions, and field of view angles of the thermal imaging and high-definition pictures are similar; the module for separating and encoding the target object is used to separate the warm objects and the inanimate background according to the different received thermal radiation energies, and encode the warm objects and the inanimate target objects respectively.

5. The working method of an unmanned aerial vehicle for identifying animals and plants according to claim 4, characterized in that, the image preprocessing module is connected to the image recognition module, and the image recognition module includes a feature extraction module and a non-feature extraction module; the feature extraction module is connected to the neural network classification module, and the neural network classification module is connected to a module for calculating the coincidence degree; the non-feature extraction module is connected to a digital information conversion module, the digital information conversion module is connected to a noise judgment module, the noise judgment module judges the digital information. When there is no noise, the noise judgment module is connected to an image recognition method based on wavelet moments, the image recognition method based on wavelet moments is connected to a BP network recognition module, the BP network recognition module is connected to a resolution test module, and the resolution test module is connected to an identification module; when there is noise, that is, when the noise cannot be ignored, the noise judgment module is connected to an infrared image recognition method based on fractal features, the infrared image recognition method based on fractal features is connected to an infrared image preprocessing module, the infrared image preprocessing module is connected to an infrared image fractal feature extraction module, and the infrared image fractal feature extraction module is connected to an identification module; the identification module is connected to a coincidence degree judgment module, the coincidence degree judgment module is connected to a result submission module; the coincidence degree judgment module is connected to a manual identification module.

6. The working method of an unmanned aerial vehicle for identifying animals and plants according to claim 1, characterized in that, In the third step mentioned above, if there is no feature extraction part, the entire image is directly converted into digital information for input, and it is judged whether there is noise; If there is noise, the infrared image recognition method based on fractal features is used to extract its fractal features for texture segmentation and target recognition, and infrared image preprocessing is performed on it to enhance its contrast, infrared image fractal feature extraction is carried out, and infrared image recognition based on neural network is performed. Through the training of the original data, the optimal weight coefficients are obtained, and thus the recognition result is obtained; If there is no noise, the image recognition method based on wavelet moments is used, and the BP network is used for recognition. After the input image is normalized, polar coordinate transformation is performed, and rotation-invariant wavelet moment feature extraction is carried out, and then it is sent to the BP network classifier for recognition, so as to obtain the recognition result; If the coincidence degree is calculated in the recognition result, the result is directly submitted. If it cannot be recognized, it is submitted to manual for recognition.

Images