Complex scene dense target classification method and system based on deep learning, and medium

Abstract

The embodiment of the invention provides a deep learning-based complex scene dense target classification method and system, and a medium, and the method comprises the steps: obtaining a collection image in real time based on an unmanned plane, carrying out the processing of the collection image, and extracting the image features; screening out target features based on the image features, and establishing a target detection frame based on the target features; target moving state information in the target detection frame is acquired, and target attitude information is analyzed; marking the target based on the target attitude information to obtain marking information, and identifying the marking information based on a deep convolution model; classifying the identification information based on the category standard information to obtain target category information, and transmitting the target category information to the terminal in real time; the unmanned aerial vehicle carries the camera to collect images in real time and analyze the moving state of the target, so that the target type is dynamically identified through the deep convolution model, the type of the dense target is accurately analyzed, the target is classified, and the analysis precision is improved.

Description

Technical Field

[0001] The present application relates to the technical field of dense target detection, and more specifically, to a method, system, and medium for classifying dense targets in complex scenarios based on deep learning. Background Art

[0002] In recent years, unmanned aerial vehicles (UAVs) have been widely used in aspects such as target acquisition and processing, and target detection technology. For example, in the field of animal protection, UAVs are used to photograph bird populations, facilitating the estimation of population numbers; in the transportation field, the current number of motor vehicles is increasing day by day, which brings certain difficulties to road condition monitoring and traffic supervision. Utilizing the characteristics of UAVs being small, flexible, and having a wide shooting range, they can be applied to the detection and tracking of vehicles on the road, thereby realizing a more convenient traffic supervision mode; they can also be used for estimating the number of people in large gathering places, etc.

[0003] In existing dense target detection methods, it is impossible to perform target annotation based on real-time collected images, and it is difficult to identify the types of targets, which affects the classification of dense targets. During the process of identifying dense targets, the error is relatively large. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to provide a method, system, and medium for classifying dense targets in complex scenarios based on deep learning. By using a UAV to carry a camera to collect images in real time and analyze the target movement state, the types of targets can be dynamically identified through a deep convolutional model, accurately analyze the categories of dense targets, classify the targets, and improve the analysis accuracy.

[0005] The embodiments of the present application also provide a method for classifying dense targets in complex scenarios based on deep learning, including:

[0006] Based on the UAV, collect images in real time, process the collected images, and extract image features;

[0007] Based on the image features, screen out target features, and establish a target detection frame based on the target features;

[0008] Obtain the target movement state information within the target detection frame, and analyze the target pose information according to the target movement state information;

[0009] Based on the target pose information, annotate the target to obtain annotation information, construct a deep convolutional model, and identify the annotation information based on the deep convolutional model to obtain identification information;

[0010] Based on the category standard information, classify the identification information to obtain target category information, and transmit the target category information to the terminal in real time.

[0011] Optionally, in the complex scene dense target classification method based on deep learning described in the embodiments of the present application, images are collected in real time based on a drone, the collected images are processed, and image features are extracted, specifically including:

[0012] Set the flight parameters of the drone, and carry a camera to collect images in real time based on the drone flight parameters to obtain a number of collected images;

[0013] Perform pixel analysis and similarity analysis on a number of collected images to obtain duplicate images and similar images;

[0014] Eliminate the duplicate images and images with a similarity greater than the set similarity threshold to obtain valid images;

[0015] Perform enhancement processing on the valid images to obtain enhanced images, and extract the features of the enhanced images to obtain image features.

[0016] Optionally, in the complex scene dense target classification method based on deep learning described in the embodiments of the present application, target features are screened out based on the image features, and a target detection frame is established based on the target features, specifically including:

[0017] Obtain the image features, and compare the image features with a set of multiple feature intervals, where the multiple feature intervals include a first feature interval, a second feature interval, and a third feature interval;

[0018] Determine the feature interval where the image features are located;

[0019] If the image features are in the first feature interval, the image features are determined as background features. If the image features are in the second feature interval, the image features are determined as fixed object features. If the image features are in the third feature interval, the image features are determined as moving object features;

[0020] Use the moving object features as the target features, analyze the size and shape of the target based on the target features, and establish a target detection frame.

[0021] Optionally, in the complex scene dense target classification method based on deep learning described in the embodiments of the present application, obtain the target movement state information within the target detection frame, and analyze the target pose information according to the target movement state information, specifically including:

[0022] Obtain the target detection frame, calculate the size of the target detection frame, and generate a detection area based on the size of the target detection frame;

[0023] Obtain the target movement state information within the detection area, where the target movement state information includes the target movement speed, the target movement direction, and the target position change information;

[0024] Analyze the target attitude information based on the target moving speed, target moving direction, and target position change information.

[0025] Optionally, in the method for classifying dense targets in complex scenarios based on deep learning described in the embodiments of the present application, after obtaining the target attitude information, the following steps are further included:

[0026] Obtain the target attitude information, based on the target attitude information;

[0027] Compare the target attitude information with the set attitude information to obtain the attitude deviation rate;

[0028] Determine whether the attitude deviation rate is greater than or equal to the set deviation rate threshold;

[0029] If it is greater than or equal to, generate correction information, and correct the target attitude information based on the correction information;

[0030] If it is less than, analyze the target tilt angle, and rotate the image based on the target tilt angle.

[0031] Optionally, in the method for classifying dense targets in complex scenarios based on deep learning described in the embodiments of the present application, label the target based on the target attitude information to obtain label information, construct a deep convolutional model, and identify the label information based on the deep convolutional model to obtain identification information, specifically including:

[0032] Obtain sample data of moving objects of several different categories, and set the proportion of the sample data of moving objects of different categories according to the category;

[0033] Based on the set proportion, select the amount of sample data of moving objects of different categories to establish a training set;

[0034] Iteratively train the model based on the training set to generate a deep convolutional model;

[0035] Obtain the target attitude information, label different categories of targets based on the target attitude information to obtain label information of multiple categories;

[0036] Input the label information into the deep convolutional model, and output the identification information.

[0037] In a second aspect, an embodiment of the present application provides a system for classifying dense targets in complex scenarios based on deep learning. The system includes: a memory and a processor. The memory includes a program of the method for classifying dense targets in complex scenarios based on deep learning. When the program of the method for classifying dense targets in complex scenarios based on deep learning is executed by the processor, the following steps are implemented:

[0038] Based on the drone, obtain the captured image in real time, process the captured image, and extract image features;

[0039] Filter out target features based on image features and establish a target detection frame based on the target features;

[0040] Obtain the target movement status information within the target detection frame and analyze the target pose information according to the target movement status information;

[0041] Annotate the target based on the target pose information to obtain annotation information, construct a deep convolutional model, and identify the annotation information based on the deep convolutional model to obtain identification information;

[0042] Classify the identification information based on the category standard information to obtain the target category information and transmit the target category information to the terminal in real time.

[0043] Optionally, in the complex scene dense target classification system based on deep learning described in the embodiments of the present application, collect images in real time based on a drone, process the collected images, and extract image features, specifically including:

[0044] Set the drone flight parameters, and carry a camera to collect images in real time based on the drone flight parameters to obtain a plurality of collected images;

[0045] Perform pixel analysis and similarity analysis on the plurality of collected images to obtain duplicate images and similar images;

[0046] Eliminate the duplicate images and the images with a similarity greater than the set similarity threshold to obtain valid images;

[0047] Perform enhancement processing on the valid images to obtain enhanced images, and extract the features of the enhanced images to obtain image features.

[0048] Optionally, in the complex scene dense target classification system based on deep learning described in the embodiments of the present application, filter out target features based on image features and establish a target detection frame based on the target features, specifically including:

[0049] Obtain image features, compare the image features with a set of multiple feature intervals, and the multiple feature intervals include a first feature interval, a second feature interval, and a third feature interval;

[0050] Judge the feature interval where the image features are located;

[0051] If the image features are in the first feature interval, the image features are determined to be background features. If the image features are in the second feature interval, the image features are determined to be fixed object features. If the image features are in the third feature interval, the image features are determined to be moving object features;

[0052] Use the moving object features as the target features, analyze the size and shape of the target based on the target features, and establish a target detection frame.

[0053] In a third aspect, an embodiment of the present application further provides a computer-readable storage medium, which includes a program for a method of classifying dense targets in complex scenarios based on deep learning. When the program for the method of classifying dense targets in complex scenarios based on deep learning is executed by a processor, the steps of the method of classifying dense targets in complex scenarios based on deep learning as described in any one of the above are implemented.

[0054] As can be seen from the above, a method, a system and a medium for classifying dense targets in complex scenarios based on deep learning provided by an embodiment of the present application acquire and collect images in real time based on an unmanned aerial vehicle (UAV), process the collected images, and extract image features; screen out target features based on the image features, and establish a target detection frame based on the target features; obtain the target movement state information within the target detection frame, and analyze the target pose information according to the target movement state information; label the target based on the target pose information to obtain labeling information, construct a deep convolutional model, and identify the labeling information based on the deep convolutional model to obtain identification information; classify the identification information based on the category standard information to obtain target category information, and transmit the target category information to the terminal in real time; collect images in real time by the UAV carrying a camera, and analyze the target movement state, so as to dynamically identify the target types through the deep convolutional model, accurately analyze the categories of dense targets, classify the targets, and improve the analysis accuracy. Description of the Drawings

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0056] Figure 1 It is a flowchart of the method for classifying dense targets in complex scenarios based on deep learning provided by an embodiment of the present application;

[0057] Figure 2 It is a flowchart of the method for processing the collected images of the method for classifying dense targets in complex scenarios based on deep learning provided by an embodiment of the present application;

[0058] Figure 3 It is a flowchart of the method for generating a target detection frame of the method for classifying dense targets in complex scenarios based on deep learning provided by an embodiment of the present application. Detailed Embodiments

[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application claimed, but only represents the selected embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative efforts belong to the scope of protection of the present application.

[0060] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of the present application, the terms "first", "second", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0061] Please refer to Figure 1 , Figure 1 which is a flowchart of a method for classifying dense targets in complex scenarios based on deep learning in some embodiments of the present application. The method for classifying dense targets in complex scenarios based on deep learning is used in a terminal device. The method for classifying dense targets in complex scenarios based on deep learning includes the following steps:

[0062] S101, obtain acquisition images in real time based on a drone, process the acquisition images, and extract image features;

[0063] S102, screen out target features based on the image features, and establish a target detection box based on the target features;

[0064] S103, obtain the target movement state information within the target detection box, and analyze the target pose information according to the target movement state information;

[0065] S104, label the target based on the target pose information to obtain labeling information, construct a deep convolutional model, and identify the labeling information based on the deep convolutional model to obtain identification information;

[0066] S105, classify the identification information based on the category standard information to obtain target category information, and transmit the target category information to the terminal in real time.

[0067] It should be noted that by collecting images in real time, analyzing the image features to obtain target features, and identifying the categories of the target features according to the deep convolutional model, the dense targets are classified, thereby improving the classification accuracy.

[0068] Please refer toFigure 2 , Figure 2 This is a flow chart of a method for processing collected images based on a deep learning-based complex scene dense target classification method in some embodiments of the present application. According to an embodiment of the present invention, based on the real-time acquisition of collected images by drones, the collected images are processed and image features are extracted, specifically including:

[0069] S201, setting the flight parameters of the UAV, and carrying a camera to collect images in real time based on the flight parameters of the UAV to obtain a number of collected images;

[0070] S202, performing pixel analysis and similarity analysis on a plurality of collected images to obtain repeated images and similar images;

[0071] S203, removing duplicate images and images whose similarity is greater than a set similarity threshold to obtain valid images;

[0072] S204, performing enhancement processing on the effective image to obtain an enhanced image, and extracting features of the enhanced image to obtain image features.

[0073] It should be noted that the image is collected by setting the UAV's flight speed, flight direction and camera acquisition frequency, and the collected images are processed to eliminate duplicate images and blurred images. After elimination, the effective images are enhanced to improve the resolution of the effective images.

[0074] Please refer to Figure 3 , Figure 3 This is a flow chart of a method for generating a target detection frame based on a complex scene dense target classification method based on deep learning in some embodiments of the present application. According to an embodiment of the present invention, target features are screened out based on image features, and a target detection frame is established based on the target features, specifically including:

[0075] S301, acquiring image features, and comparing the image features with a plurality of set feature intervals, where the plurality of feature intervals include a first feature interval, a second feature interval, and a third feature interval;

[0076] S302, determining the feature interval in which the image feature is located;

[0077] S303, if the image feature is in the first feature interval, the image feature is determined as a background feature; if the image feature is in the second feature interval, the image feature is determined as a fixed object feature; if the image feature is in the third feature interval, the image feature is determined as a moving object feature;

[0078] S304, taking the moving object features as target features, analyzing the size and shape of the target based on the target features, and establishing a target detection frame.

[0079] It should be noted that by setting multiple feature intervals and analyzing the target interval in which the image features are located, the background features, moving object features, and fixed object features can be accurately screened, and then the target features can be obtained, improving the efficiency of dense target analysis.

[0080] According to an embodiment of the present invention, obtaining the target movement state information within the target detection frame and analyzing the target pose information based on the target movement state information specifically includes:

[0081] Obtaining the target detection frame, calculating the size of the target detection frame, and generating a detection area based on the size of the target detection frame;

[0082] Obtaining the target movement state information within the detection area, where the target movement state information includes the target movement speed, target movement direction, and target position change information;

[0083] Analyzing the target pose information based on the target movement speed, target movement direction, and target position change information.

[0084] It should be noted that the detection area is set according to the target detection frame, and the target movement state information is analyzed, so as to accurately analyze the target pose.

[0085] According to an embodiment of the present invention, after obtaining the target pose information, it further includes:

[0086] Obtaining the target pose information and based on the target pose information;

[0087] Comparing the target pose information with the set pose information to obtain a pose deviation rate;

[0088] Judging whether the pose deviation rate is greater than or equal to the set deviation rate threshold;

[0089] If it is greater than or equal to, generating correction information and correcting the target pose information based on the correction information;

[0090] If it is less than, analyzing the target tilt angle and rotating the image based on the target tilt angle.

[0091] It should be noted that by analyzing the target pose information, correcting the target pose, and at the same time analyzing the target tilt angle and rotating the image for adjustment, it is ensured that the target is always in the positive direction during the target pose analysis, preventing the target from tilting and improving the analysis accuracy of the target pose.

[0092] According to an embodiment of the present invention, annotating the target based on the target pose information to obtain annotation information, constructing a deep convolutional model, and identifying the annotation information based on the deep convolutional model to obtain identification information, specifically including:

[0093] Obtain sample data of moving objects of several different categories, and set the proportion of the sample data of moving objects of different categories according to the category;

[0094] Based on the set proportion, select the amount of sample data of moving objects of different categories to establish a training set;

[0095] Based on the training set, perform iterative training on the model to generate a deep convolutional model;

[0096] Obtain the target pose information, and label the targets of different categories based on the target pose information to obtain annotation information of multiple categories;

[0097] Input the annotation information into the deep convolutional model and output the recognition information.

[0098] It should be noted that by continuously training the model with the sample data of moving objects, the output result of the model is made closer to the actual result, improving the recognition accuracy.

[0099] In a second aspect, an embodiment of the present application provides a complex scene dense target classification system based on deep learning. The system includes: a memory and a processor. The memory includes a program of a complex scene dense target classification method based on deep learning. When the program of the complex scene dense target classification method is executed by the processor, the following steps are implemented:

[0100] Based on the drone, obtain the collected images in real time, process the collected images, and extract image features;

[0101] Based on the image features, screen out the target features, and establish a target detection frame based on the target features;

[0102] Obtain the target movement state information within the target detection frame, and analyze the target pose information according to the target movement state information;

[0103] Based on the target pose information, label the target to obtain annotation information, construct a deep convolutional model, and perform recognition on the annotation information based on the deep convolutional model to obtain recognition information;

[0104] Based on the category standard information, classify the recognition information to obtain the target category information, and transmit the target category information to the terminal in real time.

[0105] It should be noted that by collecting images in real time, analyzing the image features, obtaining the target features, and performing category recognition on the target features according to the deep convolutional model, the dense targets are classified, improving the classification accuracy.

[0106] According to an embodiment of the present invention, based on the drone, obtaining the collected images in real time, processing the collected images, and extracting image features specifically include:

[0107] Set the flight parameters of the drone, and based on the flight parameters of the drone, carry a camera to collect images in real time to obtain a number of collected images;

[0108] Perform pixel analysis and similarity analysis on a number of collected images to obtain duplicate images and similar images;

[0109] Eliminate the duplicate images and the images with similarity greater than the set similarity threshold to obtain valid images;

[0110] Perform enhancement processing on the valid images to obtain enhanced images, and extract the features of the enhanced images to obtain image features.

[0111] It should be noted that by setting the flight speed, flight direction of the drone and the camera acquisition frequency to collect images, and processing the collected images, eliminating duplicate images and blurred images, and enhancing the valid images after elimination to improve the resolution of the valid images.

[0112] According to the embodiment of the present invention, screening out target features based on the image features, and establishing a target detection frame based on the target features, specifically including:

[0113] Obtain the image features, and compare the image features with a set of multiple feature intervals, and the multiple feature intervals include a first feature interval, a second feature interval and a third feature interval;

[0114] Judge the feature interval where the image features are located;

[0115] If the image features are in the first feature interval, the image features are determined to be background features. If the image features are in the second feature interval, the image features are determined to be fixed object features. If the image features are in the third feature interval, the image features are determined to be moving object features;

[0116] Take the moving object features as the target features, analyze the size and shape of the target based on the target features, and establish a target detection frame.

[0117] It should be noted that by setting multiple feature intervals and analyzing the target interval where the image features are located, the background features, moving object features and fixed object features can be accurately screened, and then the target features can be obtained to improve the efficiency of dense target analysis.

[0118] According to the embodiment of the present invention, obtaining the target movement state information within the target detection frame, and analyzing the target attitude information according to the target movement state information, specifically including:

[0119] Obtain the target detection frame, calculate the size of the target detection frame, and generate a detection area based on the size of the target detection frame;

[0120] Obtain the target movement status information within the detection area, where the target movement status information includes the target movement speed, the target movement direction, and the target position change information;

[0121] Analyze the target pose information based on the target movement speed, the target movement direction, and the target position change information.

[0122] It should be noted that the detection area is set according to the target detection frame, and the target movement status information is analyzed to accurately analyze the target pose.

[0123] According to the embodiments of the present invention, after obtaining the target pose information, it further includes:

[0124] Obtain the target pose information, based on the target pose information;

[0125] Compare the target pose information with the set pose information to obtain the pose deviation rate;

[0126] Determine whether the pose deviation rate is greater than or equal to the set deviation rate threshold;

[0127] If it is greater than or equal to, generate correction information and correct the target pose information based on the correction information;

[0128] If it is less than, analyze the target tilt angle and rotate the image based on the target tilt angle.

[0129] It should be noted that by analyzing the target pose information, correcting the target pose, analyzing the target tilt angle at the same time, and rotating and adjusting the image, it is ensured that the target is always in the forward direction during target pose analysis, preventing the target from tilting and improving the analysis accuracy of the target pose.

[0130] According to the embodiments of the present invention, based on the target pose information, the target is labeled to obtain the labeling information, a deep convolutional model is constructed, and the labeling information is recognized based on the deep convolutional model to obtain the recognition information, specifically including:

[0131] Obtain the sample data of moving objects of several different categories, and set the proportion of the sample data of moving objects of different categories according to the category;

[0132] Based on the set proportion, select the sample data volume of moving objects of different categories to establish a training set;

[0133] Iteratively train the model based on the training set to generate a deep convolutional model;

[0134] Obtain the target pose information, label different categories of targets based on the target pose information to obtain the labeling information of multiple categories;

[0135] Input the labeling information into the deep convolutional model and output the recognition information.

[0136] It should be noted that by continuously training the model with the moving object sample data, the output result of the model can be made closer to the actual result, thereby improving the recognition accuracy.

[0137] The third aspect of the present invention provides a computer-readable storage medium, which includes a program for the method of classifying dense targets in complex scenes based on deep learning. When the program for the method of classifying dense targets in complex scenes based on deep learning is executed by a processor, the steps of the method of classifying dense targets in complex scenes based on deep learning as described in any one of the above are implemented.

[0138] A method, system and medium for classifying dense targets in complex scenes based on deep learning disclosed by the present invention, which includes: acquiring and collecting images in real time based on an unmanned aerial vehicle (UAV), processing the acquired images, and extracting image features; screening out target features based on the image features, and establishing a target detection frame based on the target features; acquiring the target movement state information within the target detection frame, and analyzing the target pose information according to the target movement state information; annotating the target based on the target pose information to obtain annotation information, constructing a deep convolutional model, and identifying the annotation information based on the deep convolutional model to obtain identification information; classifying the identification information based on the category standard information to obtain target category information, and transmitting the target category information to the terminal in real time; collecting images in real time by the UAV carrying a camera, and analyzing the target movement state, so as to dynamically identify the target type through the deep convolutional model, accurately analyze the categories of dense targets, classify the targets, and improve the analysis accuracy.

[0139] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the couplings, direct couplings, or communication connections between the components shown or discussed with each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be electrical, mechanical, or other forms.

[0140] The units described as separate components above may or may not be physically separated, and the components shown as units may or may not be physical units; they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0141] In addition, in each embodiment of the present invention, each functional unit can be entirely integrated into one processing unit, or each unit can be separately regarded as one unit, or two or more units can be integrated into one unit; the above-mentioned integrated unit can be implemented in the form of hardware, or in the form of a hardware plus a software functional unit.

[0142] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks and other various media that can store program codes.

[0143] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks, or optical disks and other various media that can store program codes.

Claims

1. A method for classifying dense objects in complex scenes based on deep learning, characterized in that: include: Based on the real-time acquisition of collected images by drones, the collected images are processed and image features are extracted; Filter out target features based on image features, and establish a target detection frame based on the target features; Obtain target movement state information within the target detection frame, and analyze target posture information based on the target movement state information; The target is labeled based on the target posture information to obtain the labeled information, a deep convolution model is constructed, and the labeled information is recognized based on the deep convolution model to obtain the recognized information; The identification information is classified based on the category standard information to obtain target category information, and the target category information is transmitted to the terminal in real time.

2. The method for classifying dense objects in complex scenes based on deep learning according to claim 1, characterized in that: Based on the real-time acquisition of collected images by drones, the collected images are processed and image features are extracted, including: Setting the flight parameters of the UAV, carrying a camera to collect images in real time based on the flight parameters of the UAV, and obtaining a number of collected images; Perform pixel analysis and similarity analysis on several collected images to obtain repeated images and similar images; Duplicate images and images whose similarity is greater than a set similarity threshold are eliminated to obtain valid images; The effective image is enhanced to obtain an enhanced image, and the features of the enhanced image are extracted to obtain image features.

3. The method for classifying dense objects in complex scenes based on deep learning according to claim 2, characterized in that: Filter out target features based on image features, and establish target detection frames based on target features, including: Acquire image features, and compare the image features with a plurality of set feature intervals, where the plurality of feature intervals include a first feature interval, a second feature interval, and a third feature interval; Determine the feature interval in which the image feature is located; If the image feature is in the first feature interval, the image feature is determined as a background feature; if the image feature is in the second feature interval, the image feature is determined as a fixed object feature; if the image feature is in the third feature interval, the image feature is determined as a moving object feature; The features of moving objects are used as target features, the size and shape of the target are analyzed based on the target features, and the target detection frame is established.

4. The method for classifying dense objects in complex scenes based on deep learning according to claim 3, characterized in that: Obtain the target movement state information within the target detection frame, and analyze the target posture information based on the target movement state information, including: Get the target detection frame, calculate the size of the target detection frame, and generate the detection area based on the size of the target detection frame; Acquire target movement state information within the detection area, wherein the target movement state information includes target movement speed, target movement direction and target position change information; Analyze the target posture information based on the target movement speed, target movement direction and target position change information.

5. The method for classifying dense objects in complex scenes based on deep learning according to claim 4, characterized in that: After obtaining the target posture information, it also includes: Obtaining target posture information, based on the target posture information; Compare the target posture information with the set posture information to obtain the posture deviation rate; Determining whether the posture deviation rate is greater than or equal to a set deviation rate threshold; If it is greater than or equal to, correction information is generated, and the target posture information is corrected based on the correction information; If it is less than, the target tilt angle is analyzed and the image is rotated based on the target tilt angle.

6. The method for classifying dense objects in complex scenes based on deep learning according to claim 5, characterized in that: The target is labeled based on the target posture information to obtain the labeled information, a deep convolution model is constructed, and the labeled information is recognized based on the deep convolution model to obtain the recognized information, which specifically includes: Obtain sample data of a number of moving objects of different categories, and set the proportions of the sample data of the moving objects of different categories according to the categories; Select the amount of sample data of moving objects of different categories based on the set ratio to establish a training set; Iteratively train the model based on the training set to generate a deep convolutional model; Obtaining target posture information, and marking targets of different categories based on the target posture information to obtain marking information of multiple categories; The annotation information is input into the deep convolutional model and the recognition information is output.

7. A complex scene dense target classification system based on deep learning, characterized in that: The system includes: a memory and a processor, wherein the memory includes a program of a complex scene dense target classification method based on deep learning, and when the program of the complex scene dense target classification method based on deep learning is executed by the processor, the following steps are implemented: Based on the real-time acquisition of collected images by drones, the collected images are processed and image features are extracted; Filter out target features based on image features, and establish a target detection frame based on the target features; Obtain target movement state information within the target detection frame, and analyze target posture information based on the target movement state information; The target is labeled based on the target posture information to obtain the labeled information, a deep convolution model is constructed, and the labeled information is recognized based on the deep convolution model to obtain the recognized information; The identification information is classified based on the category standard information to obtain target category information, and the target category information is transmitted to the terminal in real time.

8. The complex scene dense target classification system based on deep learning according to claim 7, characterized in that: Based on the real-time acquisition of collected images by drones, the collected images are processed and image features are extracted, including: Setting the flight parameters of the UAV, carrying a camera to collect images in real time based on the flight parameters of the UAV, and obtaining a number of collected images; Perform pixel analysis and similarity analysis on several collected images to obtain repeated images and similar images; Duplicate images and images whose similarity is greater than a set similarity threshold are eliminated to obtain valid images; The effective image is enhanced to obtain an enhanced image, and the features of the enhanced image are extracted to obtain image features.

9. The complex scene dense target classification system based on deep learning according to claim 8, characterized in that: Filter out target features based on image features, and establish target detection frames based on target features, including: Acquire image features, and compare the image features with a plurality of set feature intervals, where the plurality of feature intervals include a first feature interval, a second feature interval, and a third feature interval; Determine the feature interval in which the image feature is located; If the image feature is in the first feature interval, the image feature is determined as a background feature; if the image feature is in the second feature interval, the image feature is determined as a fixed object feature; if the image feature is in the third feature interval, the image feature is determined as a moving object feature; The features of moving objects are used as target features, the size and shape of the target are analyzed based on the target features, and the target detection frame is established.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a complex scene dense target classification method program based on deep learning. When the complex scene dense target classification method program based on deep learning is executed by a processor, the steps of the complex scene dense target classification method based on deep learning as described in any one of claims 1 to 6 are implemented.

Images