A method and system for identifying fast-growing tree species in transmission line channels

Abstract

The present invention discloses a method and system for identifying fast-growing tree species in a transmission line channel, which specifically include the following steps: using a rotating target detection algorithm to identify the position of a pole tower in a panoramic puzzle; delineating a transmission line buffer zone according to the position of the pole tower; using semantic segmentation to identify the fast-growing tree species in the transmission line buffer zone; generating a file in a spatial data open format representing the distribution of fast-growing tree species. The present invention is based on a transmission line channel drone inspection panoramic puzzle, adopts an optimized OrientedR‑CNN model to obtain the position of the pole tower through a rotating target detection algorithm, optimizes the existing UperNet model in view of the characteristics of fast-growing trees with uncertain size and shape and relatively simple features, so as to more accurately identify fast-growing tree species, improve work efficiency and accuracy, and overcome the problem of low efficiency in the original manual analysis of tree species hidden dangers in transmission lines.

Description

Technical Field

[0001] The present invention relates to the technical fields of power facility inspection, tree species identification, and the like, and in particular to a method and system for identifying fast-growing tree species in a transmission line channel. Background Art

[0002] In recent years, the country has attached increasing importance to environmental protection, and forestry for economic purposes and environmental protection has also developed rapidly, which has brought hidden dangers to the transmission line channels. When the trees under the transmission lines grow to a certain height, they may cause line failures, affecting the safety and stability of the entire transmission line. According to relevant reliable statistics, tree barriers account for 50%~60% of the cases where transmission line safety hazards lead to failures, and the rest, such as illegal construction and buildings, account for only 10%~15%. It can be seen that tree barriers have the greatest impact on the safety hazards of transmission lines.

[0003] Fast-growing tree species are the most harmful due to their fast growth and high growth height. Therefore, it is necessary to grasp the distribution of fast-growing tree species under the transmission lines in real time and deal with them. With the rapid development and maturity of drone technology, the use of drones to assist in the identification of tree species under transmission lines has become a reliable method. Drones take a large number of aerial photos when inspecting transmission channels. After synthesizing these photos, an oversized aerial panoramic puzzle in the format of a labeled image file can be obtained. At present, manual analysis of tree species hidden dangers in transmission lines is still the main means, but it requires more forestry expertise, is very inefficient, and has low recognition accuracy when the workload is large.

[0004] A Chinese patent with publication number CN110082780A discloses "a method for collecting tree barrier information of overhead power transmission lines". The method obtains the offline distance and offline angle between a UAV and the overhead power transmission line, the distance between the UAV and the tree barrier and the angle between the UAV and the tree barrier, and controls the UAV to follow the overhead power transmission line according to the offline distance and offline angle. The line-tree distance between the overhead power transmission line and the tree barrier is calculated according to the offline distance, offline angle, distance from the tree and angle from the tree to collect the tree barrier information. However, a large amount of data is obtained during the calculation process, and the background operation is complicated. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a method and system for identifying fast-growing tree species in a transmission line channel. Based on the panoramic puzzle of drone inspection of the transmission line channel, it is proposed to use a rotating target detection algorithm to obtain the position of the tower. In view of the characteristics of fast-growing trees with irregular sizes and shapes and relatively simple features, the existing UperNet model is optimized, which can more accurately identify fast-growing tree species and improve work efficiency and accuracy.

[0006] The technical solution of the present invention is as follows:

[0007] A method for identifying fast-growing tree species in a transmission line channel, comprising the following steps:

[0008] S1. Obtain a panoramic jigsaw puzzle of a power transmission line channel, preprocess the panoramic jigsaw puzzle of the power transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion;

[0009] S2, constructing a pole tower target detection model, inputting the first slice training set into the pole tower target detection model for iterative training, obtaining a trained pole tower target detection model after N iterative trainings, and inputting the first slice test set into the pole tower target detection model for performance testing;

[0010] S3, inputting the panoramic puzzle to be detected into the pole tower target detection model to obtain the pole tower position, delineating the transmission line buffer zone according to the pole tower position, preprocessing the transmission line buffer zone to obtain a second slice data set, and dividing the second slice data set into a second slice training set and a second slice test set in proportion;

[0011] S4, constructing a fast-growing tree species recognition model, inputting the second slice training set into the fast-growing tree species recognition model for iterative training, obtaining a trained fast-growing tree species recognition model after N iterative trainings, and inputting the second slice test set into the fast-growing tree species recognition model for performance testing;

[0012] S5. Input the panorama puzzle to be detected into the fast-growing tree species recognition model to generate a file in an open spatial data format representing the distribution of fast-growing tree species.

[0013] Furthermore, the step S1 is specifically as follows: dividing the transmission line channel panoramic puzzle to obtain a first slice whose size and overlap rate are within a preset threshold range.

[0014] Furthermore, the tower target detection model in step S2 is an optimized OrientedR-CNN model, wherein the specific steps of optimizing the OrientedR-CNN model are: expanding the aspect ratio and scale of the preset anchor box, the optimized AnchorScales is 8, and the AnchorRatios is [0.5, 1, 2]; the img_size parameter of the model is set to 1024, the optimizer is set to AdamW, betas is set to (0.9, 0.999), lr is set to 0.0001, and weight_decay is set to 0.05; backbone is set to SwinTransformer-tiny; the normalization parameter mean is [101.319, 120.211, 108.69], and the std is [36.037, 36.945, 42.36]; the long edge definition method is used to define the rotation box.

[0015] Furthermore, the step S3 also includes converting the detected transmission line channel panoramic puzzle into a global pixel coordinate system, using global pixel coordinates to represent the tower position in the global pixel coordinate system, and post-processing the detected panoramic puzzle using non-maximum suppression.

[0016] Furthermore, the transmission line buffer in step S3 is delineated according to the center point coordinates of the towers and the voltage level, specifically: reading the center point pixel coordinates of two adjacent towers in the global pixel coordinate system, and setting a rectangle with the center point line of the two adjacent towers as the axis of symmetry, wherein the length of the rectangle is the center point distance of the two adjacent towers, and the width of the rectangle corresponds to the weight of the voltage level.

[0017] Furthermore, the step S4 is specifically: dividing the transmission line buffer zone to obtain second slices whose sizes and overlap rates are within a preset threshold range.

[0018] Furthermore, the fast-growing tree species recognition model in step S5 is an optimized UperNet model, wherein the specific steps of optimizing the UperNet model are: concatenating the last two layers of fpn in UperNet as the input of fpn_bottleneck. The number of input channels of fpn_bottleneck is 2*512, the number of output channels is 512, the convolution kernel size is set to 3, the stride is 1, and the padding is 1; the loss_weight of UperNet_head is set to 0.9; the normalization parameter mean is [101.319 , 120.211 , 108.69], and the std is [36.037 , 36.945 , 42.36]; the image_size of the model is set to 672; the ratio_range of Resize in data enhancement is set to (0.7,1.1), and the backbone is set to SwinTransformer-small; auxiliary_head is set to FCNHead, in_channels is set to 512, channels is set to 256, dropout_ratio is set to 0.15, and loss_weight is set to 0.5.

[0019] Furthermore, step S5 also includes if the results of identifying fast-growing tree species in the overlapping parts of two adjacent second slices are different, then re-slicing to identify the fast-growing tree species with the overlapping parts of the two adjacent second slices with different fast-growing tree species identification results as the center, and using the identified overlapping parts as the final identification result.

[0020] A transmission line channel fast-growing tree species identification system, the transmission line channel fast-growing tree species identification system comprises a first data acquisition module, a first data processing module, a second data acquisition module, and a second data processing module, wherein:

[0021] The first data acquisition module is used to obtain a panoramic mosaic of the transmission line channel, pre-process the panoramic picture of the transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion and transmit them to the first data processing module;

[0022] The first data processing module is equipped with a pole tower target detection model, which is trained and tested by inputting a first slice training set and a first slice test set, and inputting a panoramic puzzle of the transmission line channel to be detected into the trained pole tower target detection model to obtain a detection result; the detection result is post-processed by a non-maximum suppression method to obtain the pole tower position, and the transmission line buffer zone is delineated according to the pole tower position;

[0023] The second data acquisition module is used to obtain a transmission line panoramic jigsaw puzzle with a transmission line buffer zone delineated, segment the transmission line panoramic picture with the transmission line buffer zone delineated to obtain a second slice data set, and divide the second slice data set into a second slice training set and a second slice test set according to proportion; the second data processing module;

[0024] The second data processing module is equipped with a fast-growing tree species recognition model, which is trained and tested by inputting a second slice training set and a second slice test set, and inputting the panoramic puzzle of the transmission line channel to be detected into the trained fast-growing tree species recognition model to perform fast-growing tree recognition and obtain a tree species recognition result.

[0025] An electronic device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that when the processor executes the program, the method for identifying fast-growing tree species in a transmission line channel is implemented.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] The present invention is based on the panoramic puzzle of transmission line channel drone inspection, and adopts an optimized pole tower target detection model, specifically an OrientedR-CNN model, which obtains the position of the pole tower in the panoramic puzzle through a rotating target detection algorithm, and optimizes the existing UperNet model in view of the characteristics of fast-growing trees with uncertain size and shape and relatively simple features, so as to more accurately identify the fast-growing tree species in the panoramic puzzle, timely analyze the hidden dangers of tree species in the transmission line, improve work efficiency and accuracy, prevent the fast-growing tree species from causing faults to the transmission line due to rapid growth, and overcome the problem of low efficiency of the original manual analysis of tree species hidden dangers in the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the structure of a method according to an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of the system structure of Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0030] In order to make the contents of the present invention easier to understand, the technical solution of the present invention is further described below in conjunction with specific implementation methods and drawings, but the present invention is not limited thereto.

[0031] Embodiment 1

[0032] See also Figure 1 The invention provides a technical solution: a method for identifying fast-growing tree species in a transmission line channel, which specifically includes the following steps:

[0033] S1. Obtain a panoramic puzzle of a transmission line channel, preprocess the panoramic puzzle of the transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion;

[0034] S2, constructing a pole tower target detection model, inputting the first slice training set into the pole tower target detection model for iterative training, obtaining a trained pole tower target detection model after N iterative trainings, and inputting the first slice test set into the pole tower target detection model for performance testing;

[0035] S3, inputting the panoramic puzzle to be detected into the pole tower target detection model to obtain the pole tower position, delineating the transmission line buffer zone according to the pole tower position, preprocessing the transmission line buffer zone to obtain a second slice data set, and dividing the second slice data set into a second slice training set and a second slice test set in proportion;

[0036] S4, constructing a fast-growing tree species recognition model, inputting the second slice training set into the fast-growing tree species recognition model for iterative training, obtaining a trained fast-growing tree species recognition model after N iterative trainings, and inputting the second slice test set into the fast-growing tree species recognition model for performance testing;

[0037] S5. Input the panorama puzzle to be detected into the fast-growing tree species recognition model to generate a file in the spatial data open format (.shp) representing the distribution of fast-growing tree species.

[0038] It should be noted that the above steps only represent the specific steps used in the implementation of this embodiment, and the order thereof is not intended to limit the technical features of the solution of the present invention.

[0039] Among them, shp is a file in the "Spatial Data Open" format. The full name of shp is "ESRI Shapefile", which is a spatial data open format developed by the Environmental Systems Research Institute of the United States. It is a vector graphics format. This format file is mainly used to describe geometric objects, including points, polylines and polygons, and can save the location and related attributes of geometric figures.

[0040] Specifically, step S1 is as follows: dividing the panoramic puzzle of the transmission line channel into first slices with a size of 1024*1024 and an overlap rate of 20%.

[0041] Specifically, the tower target detection model in step S2 is an optimized Oriented R-CNN model. Oriented R-CNN is an effective and simple oriented object detection framework, a general two-stage rotation target detector, and has good accuracy and efficiency. The optimized Oriented R-CNN model is specifically:

[0042] The backbone is set to SwinTransformer-tiny, with drop_path_rate of 0.15 and out_indices of (0, 1, 2, 3); the neck is FPN, in_channels is set to [96, 192, 384, 768], out_channels is set to 256, and num_outs is set to 5; in OrientedRPNHead, in_channels is 256, the scales of anchor_generator is set to 8, the ratios are set to [0.5, 1.0, 2.0], the strides are set to [4, 8, 16, 32, 64], the type of bbox_coder is set to MidpointOffsetCoder, the angle_range is set to le90, the target_means is set to [0.0, 0.0, 0.0, 0.0, 0.0], and the target_stds is set to [1.0, 1.0, 1.0, 1.0, 0.5, 0.5], loss_cls uses CrossEntropyLoss, loss_weight is set to 1.0, loss_bbox uses SmoothL1Loss, beta is set to 0.1111111111111111, loss_weight is set to 1.0; in the bbox_roi_extractor of OrientedStandardRoIHead, roi_layer uses RoIAlignRotated, out_size is set to 7, sample_num is set to 2, out_channels is set to 256, featmap_strides is set to [4, 8, 16, 32], bbox_head uses RotatedShared2FCBBoxHead, in_channels is set to 256, fc_out_channels is set to 1024, roi_feat_size is set to 7, and num_classes is set to 2.bbox_coder uses DeltaXYWHAOBBoxCoder, angle_range is set to le90, target_means is set to (0.0, 0.0, 0.0, 0.0, 0.0), target_stds is set to (0.1, 0.1, 0.2,0.2, 0.1), loss_cls uses CrossEntropyLoss, loss_weight is set to 1.0, loss_bbox uses SmoothL1Loss, beta is set to 1.0, loss_weight is set to 1.0; the img_size parameter of the model is set to 1024, the optimizer is set to AdamW, betas is set to (0.9, 0.999), lr is set to 0.0001, weight_decay is set to 0.05; the normalization parameter mean is [101.319 , 120.211 , 108.69], and std is [36.037 , 36.945 , 42.36].

[0043] The above parameter settings can effectively improve the accuracy of tower recognition, and can be used on the panoramic puzzle data obtained from the current transmission line inspection to achieve relatively good results.

[0044] Specifically, step S3 also includes converting the detected transmission line channel panoramic puzzle into a global pixel coordinate system, using the global pixel coordinates to represent the tower position in the global pixel coordinate system, and post-processing the detected panoramic puzzle using non-maximum suppression (NMS).

[0045] Specifically, in step S3, the transmission line buffer is delineated according to the coordinates of the center point of the tower and the voltage level, specifically: read the pixel coordinates (x1, y1) and (x2, y2) of the center points of two adjacent towers in the global pixel coordinate system, and set a rectangle with the line connecting the center points of the two adjacent towers as the symmetry axis, wherein the length of the rectangle is the distance between the center points of the two adjacent towers, and the width of the rectangle corresponds to the weight of the voltage level, which is 20*V, and V is the weight corresponding to the voltage level.

[0046] Among them, the voltage level weights are as follows: 110kV is 2.5; 220kV is 3; 330kV is 4; 500kV is 5; 750kV is 8; 1000kV is 10; the voltage levels are entered manually.

[0047] Specifically, step S4 is as follows: dividing the transmission line buffer into a plurality of second slices with a size of 672*672 and an overlap rate of 50%.

[0048] Specifically, the fast-growing tree species recognition model in step S5 is a semantic segmentation model, using the optimized UperNet model. The optimized Oriented R-CNN model is specifically:

[0049] The backbone is set to SwinTransformer-small, with drop_path_rate of 0.35 and out_indices of (0, 1, 2, 3); the decode_head uses the improved UperNetHead, in_channels is set to [96,192, 384, 768], pool_scales is set to (1, 2, 3, 6), channels is set to 512, dropout_ratio is set to 0.15, num_classes is set to 3, loss_decode uses CrossEntropyLoss, and loss_weight is set to 0.9. The input of fpn_bottleneck is the last two layers of fpn, in_channels is set to 2*512, out_channels is set to 512, kernel_size is set to 3, stride is 1, padding is 1; auxiliary_head is set to FCNHead, in_channels is set to 384, channels is set to 256, num_convs is set to 1, dropout_ratio is set to 0.05, num_classes is set to 4, loss_decode uses CrossEntropyLoss, and loss_weight is set to 0.5. The normalization parameter mean is [101.319, 120.211, 108.69], and std is [36.037, 36.945, 42.36]; the image_size of the model is set to 672; the ratio_range of Resize in data enhancement is set to (0.7, 1.1).

[0050] The above parameter settings can effectively improve the accuracy of fast-growing tree species recognition, and can be used on the panoramic puzzle data obtained from the current transmission line inspection to achieve relatively good results.

[0051] Specifically, step S5 also includes, if the recognition results of the overlapping parts of two adjacent second slices are different, re-slicing to identify the fast-growing tree species with the overlapping parts of the two adjacent second slices with different fast-growing tree species recognition results as the center, and the specific slicing method is: expanding in the length direction, expanding 118 pixels in each direction, and finally obtaining a slice with a size of 672*672, and using the identified overlapping parts as the final fast-growing tree species recognition results.

[0052] Specifically, the median value of the recognition result is 0-2, representing different types, and 0-2 represents background, bamboo, and eucalyptus respectively.

[0053] Specifically, according to the fast-growing tree species identification result, a spatial data open format file representing the distribution of the fast-growing tree species is generated, and the spatial data open format file and the panoramic puzzle in the tagged image file format have the same geographic coordinate information.

[0054] The above method provided in this embodiment can be stored in a computer-readable storage medium in a coded form and implemented in the form of a computer program, and the basic parameter information required for calculation is input through computer hardware, and the calculation result is output.

[0055] Embodiment 2

[0056] A transmission line channel fast-growing tree species identification system, the transmission line channel fast-growing tree species identification system comprises a first data acquisition module, a first data processing module, a second data acquisition module, and a second data processing module, wherein:

[0057] The first data acquisition module is used to obtain a panoramic mosaic of the transmission line channel, pre-process the panoramic image of the transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion and transmit them to the first data processing module;

[0058] The first data processing module is equipped with a pole tower target detection model, which is trained and tested by inputting the first slice training set and the first slice test set, and inputting the panoramic puzzle of the transmission line channel to be detected into the trained pole tower target detection model to obtain the detection result; the non-maximum suppression method is used to post-process the detection result to obtain the pole tower position, and the transmission line buffer zone is delineated according to the pole tower position;

[0059] The second data acquisition module is used to obtain a transmission line panoramic jigsaw puzzle with a transmission line buffer zone demarcated, segment the transmission line panoramic picture with the transmission line buffer zone demarcated to obtain a second slice data set, and divide the second slice data set into a second slice training set and a second slice test set according to proportion; the second data processing module;

[0060] The second data processing module is equipped with a fast-growing tree species recognition model. The second slice training set and the second slice test set are input for training and testing. The panoramic puzzle of the transmission line channel to be detected is input into the trained fast-growing tree species recognition model for fast-growing tree recognition to obtain the tree species recognition result.

[0061] Embodiment 3

[0062] The present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a method for identifying fast-growing tree species in a transmission line channel as described in Embodiment 1 of the present invention is implemented.

[0063] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0064] The present invention is described with reference to flowcharts of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process in the flowchart, as well as the combination of processes in the flowchart, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0065] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple functions specified in a flowchart.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 The steps of a specified function in a process or multiple processes.

[0067] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for identifying fast-growing tree species in a transmission line channel, characterized in that: The specific steps include: S1. Obtain a panoramic jigsaw puzzle of a power transmission line channel, preprocess the panoramic jigsaw puzzle of the power transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion; S2, constructing a pole tower target detection model, inputting the first slice training set into the pole tower target detection model for iterative training, obtaining a trained pole tower target detection model after N iterative trainings, and inputting the first slice test set into the pole tower target detection model for performance testing; S3, inputting the panoramic puzzle to be detected into the pole tower target detection model to obtain the pole tower position, delineating the transmission line buffer zone according to the pole tower position, preprocessing the transmission line buffer zone to obtain a second slice data set, and dividing the second slice data set into a second slice training set and a second slice test set in proportion; S4, constructing a fast-growing tree species recognition model, inputting the second slice training set into the fast-growing tree species recognition model for iterative training, obtaining a trained fast-growing tree species recognition model after N iterative trainings, and inputting the second slice test set into the fast-growing tree species recognition model for performance testing; S5, inputting the panorama puzzle to be detected into a fast-growing tree species recognition model to generate a file in an open spatial data format representing the distribution of fast-growing tree species; The tower target detection model in step S2 is an optimized OrientedR-CNN model, wherein the specific steps of optimizing the OrientedR-CNN model are as follows: the aspect ratio and scale of the preset anchor frame are expanded, and the optimized AnchorScales is 8, and the AnchorRatios is [0.5, 1, 2]; the img_size parameter of the model is set to 1024, the optimizer is set to AdamW, betas is set to (0.9, 0.999), lr is set to 0.0001, and weight_decay is set to 0.05; backbone is set to SwinTransformer-tiny; the normalization parameter mean is [101.319, 120.211, 108.69], and the std is [36.037, 36.945, 42.36]; the long side definition method is used to define the rotation frame; The fast-growing tree species recognition model in step S5 is an optimized UperNet model, wherein the specific steps of UperNet model optimization are: splicing the last two layers of fpn in UperNet as the input of fpn_bottleneck; the number of fpn_bottleneck input channels is 2*512, the number of output channels is 512, the convolution kernel size is set to 3, the stride is 1, and the padding is 1; the loss_weight of UperNet_head is set to 0.9; the normalization parameter mean is [101.319, 120.211, 108.69], and the std is [36.037, 36.945, 42.36]; the image_size of the model is set to 672; the ratio_range of Resize in data enhancement is set to (0.7, 1.1), the backbone is set to SwinTransformer-small; the auxiliary_head is set to FCNHead, in_channels is set to 512, channels is set to 256, dropout_ratio is set to 0.15, and the loss_weight is set to 0.

5.

2. A method for identifying fast-growing tree species in a transmission line channel according to claim 1, characterized in that: The step S1 specifically includes: dividing the panoramic mosaic of the transmission line channel to obtain a first slice whose size and overlap rate are within a preset threshold range.

3. A method for identifying fast-growing tree species in a transmission line channel according to claim 1, characterized in that: The step S3 also includes converting the detected transmission line channel panoramic puzzle into a global pixel coordinate system, using global pixel coordinates to represent the tower position in the global pixel coordinate system, and post-processing the detected panoramic puzzle using non-maximum suppression.

4. A method for identifying fast-growing tree species in a transmission line channel according to claim 3, characterized in that: The transmission line buffer in step S3 is delineated according to the center point coordinates of the towers and the voltage level, specifically: reading the center point pixel coordinates of two adjacent towers in the global pixel coordinate system, and setting a rectangle with the center point line of the two adjacent towers as the axis of symmetry, wherein the length of the rectangle is the center point distance of the two adjacent towers, and the width of the rectangle corresponds to the weight of the voltage level.

5. The method for identifying fast-growing tree species in a power transmission line channel according to claim 1, characterized in that: The step S4 specifically includes: dividing the transmission line buffer zone to obtain second slices whose sizes and overlap rates are within a preset threshold range.

6. A method for identifying fast-growing tree species in a transmission line channel according to claim 1, characterized in that: The step S5 also includes if the results of identifying fast-growing tree species in the overlapping parts of two adjacent second slices are different, re-slicing to identify fast-growing tree species with the overlapping parts of the two adjacent second slices having different fast-growing tree species identification results as the center, and using the identified overlapping parts as the final identification result.

7. A system for identifying fast-growing tree species in transmission line channels, characterized in that: The transmission line channel fast-growing tree species identification system comprises a first data acquisition module, a first data processing module, a second data acquisition module, and a second data processing module, wherein: The first data acquisition module is used to obtain a panoramic mosaic of the transmission line channel, pre-process the panoramic picture of the transmission line channel to obtain a first slice data set, and divide the first slice data set into a first slice training set and a first slice test set in proportion and transmit them to the first data processing module; The first data processing module is equipped with a pole tower target detection model, which is trained and tested by inputting a first slice training set and a first slice test set, and inputting a panoramic puzzle of the transmission line channel to be detected into the trained pole tower target detection model to obtain a detection result; the detection result is post-processed by a non-maximum suppression method to obtain the pole tower position, and the transmission line buffer zone is delineated according to the pole tower position; The second data acquisition module is used to obtain a transmission line panoramic jigsaw puzzle with a transmission line buffer zone delineated, segment the transmission line panoramic picture with the transmission line buffer zone delineated to obtain a second slice data set, and divide the second slice data set into a second slice training set and a second slice test set according to proportion; the second data processing module; The second data processing module is equipped with a fast-growing tree species recognition model, which is trained and tested by inputting a second slice training set and a second slice test set, and inputting the transmission line channel panoramic puzzle to be detected into the trained fast-growing tree species recognition model to perform fast-growing tree recognition to obtain a tree species recognition result; The tower target detection model is an optimized OrientedR-CNN model, wherein the specific steps of OrientedR-CNN model optimization are: the aspect ratio and scale of the preset anchor box are expanded, the optimized AnchorScales is 8, and the AnchorRatios is [0.5, 1, 2]; the img_size parameter of the model is set to 1024, the optimizer is set to AdamW, betas is set to (0.9, 0.999), lr is set to 0.0001, and weight_decay is set to 0.05; backbone is set to SwinTransformer-tiny; the normalization parameter mean is [101.319, 120.211, 108.69], and the std is [36.037, 36.945, 42.36]; the long side definition method is used to define the rotation box; The fast-growing tree species recognition model is the optimized UperNet model. The specific steps of UperNet model optimization are as follows: concatenate the last two layers of fpn in UperNet as the input of fpn_bottleneck; the number of input channels of fpn_bottleneck is 2*512, the number of output channels is 512, the convolution kernel size is set to 3, the stride is 1, and the padding is 1; the loss_weight of UperNet_head is set to 0.9; the normalization parameter mean is [101.319, 120.211, 108.69], and the std is [36.037, 36.945, 42.36]; the image_size of the model is set to 672; the ratio_range of Resize in data enhancement is set to (0.7, 1.1), the backbone is set to SwinTransformer-small; the auxiliary_head is set to FCNHead, in_channels is set to 512, channels is set to 256, dropout_ratio is set to 0.15, and the loss_weight is set to 0.

5.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for identifying fast-growing tree species in a transmission line channel as claimed in any one of claims 1 to 6 is implemented.

Images