Power transmission channel construction scene intelligent identification method and system based on unmanned aerial vehicle aerial photography
By employing unsupervised superpixel technology and fine-grained information fusion methods, the challenge of detecting construction scenes in UAV inspection images was solved, enabling efficient and intelligent identification of power transmission channel construction scenes and improving detection accuracy and computational efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2024-02-01
- Publication Date
- 2026-07-03
AI Technical Summary
Existing technologies are insufficient to effectively identify potential hazards in power transmission channel construction scenarios. In particular, the complexity and diversity of drone inspection images lead to large intra-class differences and high inter-class similarities, making it difficult to accurately detect construction hazards with varying shapes and sizes in construction scenarios.
We employ a method based on unsupervised superpixel-guided scene semantic feature clustering and fine-grained information fusion. By clustering image features using unsupervised superpixel technology and removing redundant information, we combine backbone networks and graph convolutional networks for feature selection and fusion to achieve fine classification.
It improves the detection accuracy of power transmission channel construction scenarios, can better identify construction scenarios with varying shapes and sizes, reduces the amount of calculation, alleviates data imbalance problems, and enhances detection capabilities.
Smart Images
Figure CN118155094B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power line inspection image processing and analysis technology, specifically to an intelligent recognition method and system for power transmission channel construction scenes based on drone aerial photography. Background Technology
[0002] The power industry is the most important basic energy industry in the process of my country's national economic development. It is one of the important infrastructures to ensure the normal operation of national economic development and people's lives. In order to meet the ever-increasing demand for electricity, the construction of power grid projects and their supporting facilities in my country has been increasing, and the scale of power grid construction has also been growing. The safe, stable and reliable operation of the power grid has become an urgent need for the State Grid.
[0003] Statistical results show that construction activities such as building and quarrying near power transmission corridors pose a significant threat to the safety of power transmission lines, and are one of the main hidden dangers causing power transmission line accidents. These accidents can cause property damage to the country and society, and may even lead to casualties. The construction scene represents the area under construction, and identifying dangerous construction scenes can mitigate and prevent the impact of human activities. Therefore, it is particularly important to use drones to inspect construction scenes along power transmission corridors and promptly eliminate safety hazards.
[0004] Currently, most detection methods for construction scenarios primarily employ pixel-based semantic segmentation or object detection methods. For pixel-based semantic segmentation, the complex and varied airspace environments encountered by inspection drones during flight make it difficult to capture high-level semantic information in drone inspection images rich in ground details and possessing latent semantic connotations using only pixel features. Furthermore, due to the diversity of ground features, drone inspection images exhibit significant intra-class differences and high inter-class similarity. This means that highly similar clues (such as color, brightness, and texture features) may exist between pixels of different categories, leading to blurred pixel classifications within different semantic category regions, further increasing the difficulty of detection. For object detection methods, the variable shapes and sizes of construction scenes, along with uncertain boundaries, make it difficult to select appropriate target bounding boxes for detection. For these reasons, current methods fail to achieve ideal results in detecting potential hazards during power transmission channel construction. Summary of the Invention
[0005] To address the current problem that the detection of potential hazards during power transmission channel construction is not ideal, this application provides an intelligent identification method and system for power transmission channel construction scenes based on UAV aerial photography.
[0006] The embodiments of this application are implemented as follows:
[0007] Firstly, this application provides an intelligent identification method for power transmission channel construction scenes based on drone aerial photography, including:
[0008] Acquire the initial image and read the power transmission channel image obtained from drone aerial photography;
[0009] Scene semantic feature clustering is performed on the initial image based on unsupervised superpixel guidance;
[0010] Redundant information in the clustering is removed to obtain the scene data;
[0011] Scene classification is performed based on the fusion of fine-grained information.
[0012] In one possible implementation, the step of performing scene semantic feature clustering on the initial image based on unsupervised superpixel guidance further includes:
[0013] Perform scene classification on the initial image;
[0014] Feature clustering is performed using a superpixel-guided task.
[0015] In one possible implementation, both the classification and the superpixel guidance task are performed unsupervised.
[0016] In one possible implementation, the clustered features are all labeled with a description of a construction category scenario.
[0017] In one possible implementation, the step of removing redundant information from the clustering to obtain scene data further includes:
[0018] From the clustered features, only the labeled data is extracted to obtain the scene data, which alleviates the data imbalance problem and reduces the amount of scene data, thereby reducing the computational load of subsequent tasks.
[0019] In one possible implementation, the scene data includes pixel-level features, scene texture, and scene color.
[0020] In one possible implementation, the scene classification based on fused fine-grained information further includes:
[0021] Feature extraction involves inputting the scene data into the backbone network to extract features, resulting in fine-grained features for feature selection and directly output global features.
[0022] Feature selection utilizes fully connected layers to predict and select the category of each feature point of the fine-grained features;
[0023] Feature fusion is performed by using a graph convolutional network to obtain one-dimensional feature vectors for both global and fine-grained purposes. These two vectors are then fused and fed into a fully connected layer to complete classification prediction.
[0024] In one possible implementation, the feature selection, which utilizes a fully connected layer to predict and select the category of each feature point of the fine-grained feature, further includes:
[0025] Fully connected layers are used to predict the category of each feature point;
[0026] When the highest probability of the prediction result after normalization exponential function is greater than the set threshold, the feature point is considered to be a fine-grained feature that is helpful for classification and will be selected for subsequent fusion.
[0027] When the highest probability of the prediction result after normalization exponential function is less than the set threshold, the feature point is considered a fine-grained feature that does not help with classification and will not be selected for subsequent fusion.
[0028] In one possible implementation, the feature fusion, based on a graph convolutional network, obtains one-dimensional feature vectors for both global and fine-grained aspects, fuses the two, and inputs them into a fully connected layer to complete classification prediction, further including:
[0029] For the feature points extracted by feature selection, they are regarded as a graph structure, where nodes represent features at different spatial locations and scales;
[0030] The graph is input into a graph convolutional network, which learns the relationships between different nodes and aggregates feature points into multiple super nodes through pooling layers.
[0031] The features of these supernodes are averaged to obtain a one-dimensional feature vector for fine-grained processing.
[0032] The global features directly output by the backbone network are processed by a pooling layer to obtain a one-dimensional feature vector for the entire network.
[0033] By concatenating two one-dimensional feature vectors, fine-grained features and global features can be fused together.
[0034] The input is fed into a fully connected layer to complete the classification prediction.
[0035] Secondly, this application provides an intelligent recognition system for power transmission channel construction scenes based on UAV aerial photography, including:
[0036] The image reading module is used to acquire initial images and read images of the power transmission channel obtained by drone aerial photography;
[0037] The feature clustering module is used to perform scene semantic feature clustering on the initial image based on unsupervised superpixel guidance;
[0038] The data extraction module is used to remove redundant information from the clusters to obtain scene data;
[0039] The data classification module is used to classify scenes based on fused fine-grained information.
[0040] The technical solution provided in this application can achieve at least the following beneficial effects:
[0041] The intelligent identification method system for power transmission channel construction scenes based on UAV aerial photography provided in this application innovatively combines superpixel segmentation and unsupervised clustering techniques to extract scene-level features, helping the network to recognize and analyze the scene as a whole and improve the network's detection capabilities. Furthermore, this application adopts a fine classification technique that integrates fine-grained information, extracting fine-grained features that are helpful for classification through feature selection and fusing them with global features to improve the fine classification effect. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present application of a method for intelligent identification of power transmission channel construction scenes based on drone aerial photography;
[0044] Figure 2 This is a schematic diagram of the structure of an intelligent recognition system for power transmission channel construction scenes based on drone aerial photography, as shown in an exemplary embodiment of this application;
[0045] Figure 3 This is a schematic diagram illustrating the specific process of construction scene detection in an exemplary embodiment of this application;
[0046] Figure 4 This is a schematic diagram illustrating the feature clustering results of an exemplary embodiment of this application;
[0047] Figure 5 This is a schematic diagram of scene data illustrated in an exemplary embodiment of this application;
[0048] Figure 6 This is a schematic diagram illustrating the predicted probability distribution of different parts in an exemplary embodiment of this application;
[0049] Figure 7 This is a schematic diagram of a network framework for fine-grained scene classification illustrated in an exemplary embodiment of this application;
[0050] Figure 8 This is a schematic diagram of a feature selection process shown in an exemplary embodiment of this application;
[0051] Figure 9 This is a schematic diagram of the feature fusion process shown in an exemplary embodiment of this application.
[0052] Figure label:
[0053] 1. Image reading module; 2. Feature clustering module; 3. Data extraction module; 4. Data classification module. Detailed Implementation
[0054] To make the objectives, implementation methods and advantages of this application clearer, the exemplary implementation methods of this application will be clearly and completely described below with reference to the accompanying drawings of the exemplary embodiments of this application. Obviously, the exemplary embodiments described are only some embodiments of this application, and not all embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.
[0055] It should be noted that the brief descriptions of terms in this application are only for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.
[0056] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar or related objects or entities, and do not necessarily imply a specific order or sequence, unless otherwise specified. It should be understood that such terms are interchangeable where appropriate.
[0057] The terms “comprising” and “having”, and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device that includes a range of components is not necessarily limited to all of the components that are clearly listed, but may include other components that are not clearly listed or that are inherent to such product or device.
[0058] Before explaining the intelligent identification method for power transmission channel construction scenes based on UAV aerial photography provided in this application embodiment, the application scenarios and implementation environment of this application embodiment will be introduced first.
[0059] The power industry is the most important basic energy industry in the process of my country's national economic development. It is one of the important infrastructures to ensure the normal operation of national economic development and people's lives. In order to meet the ever-increasing demand for electricity, the construction of power grid projects and their supporting facilities in my country has been increasing, and the scale of power grid construction has also been growing. The safe, stable and reliable operation of the power grid has become an urgent need for the State Grid.
[0060] Statistical results show that construction activities such as building and quarrying near power transmission corridors pose a significant threat to the safety of power transmission lines, and are one of the main hidden dangers causing power transmission line accidents. These accidents can cause property damage to the country and society, and may even lead to casualties. The construction scene represents the area under construction, and identifying dangerous construction scenes can mitigate and prevent the impact of human activities. Therefore, it is particularly important to use drones to inspect construction scenes along power transmission corridors and promptly eliminate safety hazards.
[0061] Currently, most detection methods for construction scenarios primarily employ pixel-based semantic segmentation or object detection methods. For pixel-based semantic segmentation, the complex and varied airspace environments encountered by inspection drones during flight make it difficult to capture high-level semantic information in drone inspection images rich in ground details and possessing latent semantic connotations using only pixel features. Furthermore, due to the diversity of ground features, drone inspection images exhibit significant intra-class differences and high inter-class similarity. This means that highly similar clues (such as color, brightness, and texture features) may exist between pixels of different categories, leading to blurred pixel classifications within different semantic category regions, further increasing the difficulty of detection. For object detection methods, the variable shapes and sizes of construction scenes, along with uncertain boundaries, make it difficult to select appropriate target bounding boxes for detection. For these reasons, current methods fail to achieve ideal results in detecting potential hazards during power transmission channel construction.
[0062] Existing methods can quickly mark safety hazards around power lines by identifying construction sites in drone aerial photos through target detection methods. They can also achieve rapid and full-coverage detection and intelligent identification of long linear construction areas through drone inspection aerial photography, target detection in completed construction areas, pixel coordinate positioning of construction nodes, spatial coordinate transformation of construction nodes, construction progress identification, and construction progress analysis.
[0063] The above methods all use target detection to identify construction scenes, and the shape and size of the identified construction scenes are relatively fixed, and the categories of identified construction scenes are limited, making it difficult to detect construction scenes with varying shapes and sizes and diverse categories in power transmission channels.
[0064] Based on this, this application provides an intelligent identification method for power transmission channel construction scenes based on UAV aerial photography. It intelligently identifies potential construction hazards in power transmission channels inspected by UAVs. Through superpixel technology, similar pixels are aggregated into larger areas. By combining pixels, more meaningful scene features are constructed, capturing the overall properties of ground objects and grasping and analyzing the whole picture. This allows for the acquisition of semantic information of the scene at a higher level. Compared to a single pixel, more robust features can be extracted from superpixel blocks, generating more accurate discrimination information and enabling the detection of more construction scene categories.
[0065] On the other hand, the target detection method used in the above-mentioned patent is more suitable for detecting construction scenes with relatively fixed shapes and sizes. This solution can detect construction scenes with varying shapes and sizes by classifying superpixel blocks.
[0066] Next, the technical solutions of this application and how they solve the aforementioned technical problems will be described in detail through embodiments and in conjunction with the accompanying drawings. The embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of this application.
[0067] Figure 1 This is a flowchart illustrating an exemplary embodiment of the present application of a method for intelligent identification of power transmission channel construction scenes based on drone aerial photography.
[0068] In one exemplary embodiment, such as Figure 1 As shown, an intelligent identification method for power transmission channel construction scenes based on UAV aerial photography is provided. This method may include the following steps:
[0069] Step 100: Acquire the initial image and read the power transmission channel image obtained from the drone's aerial photography.
[0070] Step 200: Perform scene semantic feature clustering on the initial image based on unsupervised superpixel guidance.
[0071] Step 300: Remove redundant information from the clusters to obtain scene data.
[0072] Step 400: Classify the scene based on the fused fine-grained information.
[0073] Figure 3 This is a schematic diagram illustrating the specific process of construction scene detection in an exemplary embodiment of this application.
[0074] In one possible implementation, such as Figure 3As shown, the specific implementation method includes five parts: scene semantic feature clustering under unsupervised superpixel guidance, redundant information removal, scene information extraction, and scene classification based on fine-grained information.
[0075] The specific steps include:
[0076] Figure 4 This is a schematic diagram illustrating the feature clustering results of an exemplary embodiment of this application.
[0077] 1. Scene semantic feature clustering guided by unsupervised superpixels
[0078] Unsupervised superpixel-guided scene semantics are used to cluster images acquired by drones, selecting scene semantic features suspected of belonging to the construction category. This lays the foundation for subsequent classification tasks. Compared to other clustering methods, some embodiments of this application have labeled features after clustering, such as... Figure 4 As shown, the light blue labels in the four different images all represent suspected construction scenarios, thus avoiding the need for manual label selection after clustering and enabling automated extraction of scene features for specified categories, providing convenience for downstream tasks.
[0079] Meanwhile, both clustering and superpixel guidance tasks are based on unsupervised learning, so the boundaries of each scene are determined according to a unified standard based on the features of each pixel, avoiding subjective human factors and solving the problem of unclear data boundaries and difficulty in labeling.
[0080] 2. Redundant Information Removal
[0081] Since the proportion of construction category data in UAV inspection image data is relatively small, it is easy to cause data imbalance. The features after unsupervised clustering are labeled. Therefore, some embodiments of this application can extract only suspected construction category data. This can alleviate the data imbalance problem and reduce the amount of scene data, thereby reducing the computational load of subsequent tasks.
[0082] Figure 5 This is a schematic diagram of scene data illustrated in an exemplary embodiment of this application.
[0083] Three: Scene Information Extraction
[0084] After clustering and redundancy removal, some embodiments of this application can be obtained as follows: Figure 5 As shown in the scene data, it can be seen that the scene-level features under superpixels not only integrate pixel-level features, but also retain various other features such as texture and color of the scene area. The fusion of multiple features improves the subsequent scene detection capabilities.
[0085] Figure 6This is a schematic diagram illustrating the predicted probability distribution of different parts in an exemplary embodiment of this application.
[0086] 4. Scene classification that integrates fine-grained information
[0087] Because the superpixel blocks are irregularly shaped, while the classification network requires square images as input, zero-padding is needed in the remaining positions. This padding occurs in all different categories of data. If training were performed directly, such as... Figure 6 As shown, the filled part results in a flatter probability distribution, which increases the difficulty of classification.
[0088] Figure 7 This is a schematic diagram of a network framework for fine-grained scene classification illustrated in an exemplary embodiment of this application. Figure 8 This is a schematic diagram illustrating a feature selection process in an exemplary embodiment of this application. Figure 9 This is a schematic diagram of the feature fusion process shown in an exemplary embodiment of this application.
[0089] To address the above issues, classification networks need to focus on the parts of the data that are conducive to classification, i.e., the parts that can broaden the probability distribution. This requires using feature selection to extract fine-grained information from the image. The network framework designed based on this approach is as follows: Figure 7 As shown, it mainly consists of three parts: feature extraction, feature selection, and feature fusion.
[0090] (1) Feature extraction
[0091] First, the image is input into the backbone network, and multiple feature maps are extracted to enrich the features. The feature extraction path is divided into two parts: one part is used for feature selection, representing fine-grained features; the other part is directly output, representing global features. In some embodiments of this application, the ResNet network is selected as the backbone network for feature extraction.
[0092] (2) Feature selection
[0093] like Figure 8 As shown, a fully connected layer is used to predict the category of each feature point. When the highest probability of the prediction result after the normalized exponential function (softmax) is greater than a set threshold, the feature point is considered a helpful fine-grained feature and will be selected for subsequent fusion. Conversely, feature points are considered fine-grained features that are less helpful for classification and are discarded.
[0094] (3) Feature fusion
[0095] Since the feature points extracted in the feature selection process are chosen based on classification probabilities, their distribution on the feature map is relatively random. Directly concatenating and fusing these feature points may corrupt the output of the backbone model.
[0096] Therefore, the following was designed: Figure 9 The feature fusion module shown treats the extracted feature points as a graph structure, where nodes represent features at different spatial locations and scales. The graph is input into a graph convolutional network (GCNN), which learns the relationships between nodes. Pooling layers then aggregate the feature points into several supernodes. Finally, the features of these supernodes are averaged to obtain a one-dimensional feature vector. Simultaneously, a global one-dimensional feature vector is also obtained after pooling layers in the backbone network. Concatenating these two one-dimensional feature vectors completes the fusion of fine-grained and global features. Finally, the fusion is input into a fully connected layer for classification prediction.
[0097] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially as indicated, these steps are not necessarily executed in the indicated order. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps.
[0098] Corresponding to the aforementioned embodiments of the intelligent identification method for power transmission channel construction scenes based on UAV aerial photography, and employing the same technical concept, this application also provides embodiments of an intelligent identification system for power transmission channel construction scenes based on UAV aerial photography.
[0099] Figure 2 This is a schematic diagram of the structure of an intelligent recognition system for power transmission channel construction scenes based on drone aerial photography, as shown in an exemplary embodiment of this application.
[0100] In one exemplary embodiment, such as Figure 2 As shown, the intelligent recognition system for power transmission channel construction scenes based on drone aerial photography includes:
[0101] Image reading module 1 is used to acquire initial images and read power transmission channel images obtained from drone aerial photography.
[0102] Feature clustering module 2 is used to perform scene semantic feature clustering on the initial image based on unsupervised superpixel guidance;
[0103] Data extraction module 3 is used to remove redundant information from clusters to obtain scene data;
[0104] Data classification module 4 is used to classify scenes based on fused fine-grained information.
[0105] Specific limitations regarding the intelligent recognition system for power transmission channel construction scenes based on UAV aerial photography can be found in the limitations of the intelligent recognition method for power transmission channel construction scenes based on UAV aerial photography mentioned above, and will not be repeated here. Each module in the aforementioned intelligent recognition system for power transmission channel construction scenes based on UAV aerial photography can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, and their corresponding operations can be executed by the processor.
[0106] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0107] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.
Claims
1. A method for intelligent identification of power transmission corridor construction scene based on aerial photography by a UAV, characterized in that, include: Acquire initial image: Read the power transmission channel image obtained from drone aerial photography; Scene semantic feature clustering is performed on the initial image based on unsupervised superpixel guidance; Redundant information in the clustering is removed to obtain the scene data; Scene classification is performed based on the fusion of fine-grained information; The scene classification based on fused fine-grained information further includes: Feature extraction: The scene data is input into the backbone network for feature extraction to obtain fine-grained features for feature selection and global features for direct output; Feature selection: The class of each feature point of the fine-grained feature is predicted using a fully connected layer, and the fine-grained feature is selected based on the highest probability of the prediction result; Feature fusion: Based on graph convolutional networks, one-dimensional feature vectors for both global and fine-grained purposes are obtained, and these two vectors are fused and input into a fully connected layer to complete classification prediction. Further steps include: For the feature points extracted by feature selection, they are regarded as a graph structure, where nodes represent features at different spatial locations and scales; The graph is input into a graph convolutional network, which learns the relationships between different nodes and aggregates feature points into multiple super nodes through pooling layers. The features of these supernodes are averaged to obtain a one-dimensional feature vector for fine-grained processing. The global features directly output by the backbone network are processed by a pooling layer to obtain a one-dimensional feature vector for the entire network. By concatenating two one-dimensional feature vectors, fine-grained features and global features can be fused together. The input is fed into a fully connected layer to complete the classification prediction.
2. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 1, characterized in that, The step of performing scene semantic feature clustering on the initial image based on unsupervised superpixel guidance further includes: Perform scene classification on the initial image; Feature clustering is performed using a superpixel-guided task.
3. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 2, characterized in that, Both the classification and the superpixel guidance task are performed in an unsupervised manner.
4. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 2, characterized in that, The clustered features all have labels representing construction category scenarios.
5. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 1, characterized in that, The process of removing redundant information from clustering to obtain scene data further includes: From the clustered features, only the labeled data is extracted to obtain the scene data, which alleviates the data imbalance problem and reduces the amount of scene data, thereby reducing the computational load of subsequent tasks.
6. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 5, characterized in that, The scene data includes pixel-level features, scene texture, and scene color.
7. The intelligent identification method for power transmission channel construction scenes based on UAV aerial photography as described in claim 1, characterized in that, The feature selection, which uses a fully connected layer to predict and select the category of each feature point of the fine-grained feature, further includes: Fully connected layers are used to predict the category of each feature point; When the highest probability of the prediction result after normalization exponential function is greater than the set threshold, the feature point is considered to be a fine-grained feature that is helpful for classification and will be selected for subsequent fusion. When the highest probability of the prediction result after normalization exponential function is less than the set threshold, the feature point is considered a fine-grained feature that does not help with classification and will not be selected for subsequent fusion.
8. A smart identification system for power transmission channel construction scenes based on UAV aerial photography, the system being used to implement the smart identification method for power transmission channel construction scenes based on UAV aerial photography as described in any one of claims 1 to 7, characterized in that, include: The image reading module is used to acquire the initial image: reading the power transmission channel image obtained by drone aerial photography; The feature clustering module is used to perform scene semantic feature clustering on the initial image based on unsupervised superpixel guidance; The data extraction module is used to remove redundant information from the clusters to obtain scene data; The data classification module is used to classify scenes based on fused fine-grained information.