CN114626572A

CN114626572A - Power transmission line path optimization method based on intelligent image recognition

Info

Publication number: CN114626572A
Application number: CN202210079335.1A
Authority: CN
Inventors: 李伟; 刘新臣; 郎垚; 蒋伟; 刘力; 文康; 刘玉然; 肖健一; 魏星; 成聪
Original assignee: Sichuan Electric Power Design and Consulting Co Ltd
Current assignee: Sichuan Electric Power Design and Consulting Co Ltd
Priority date: 2022-01-24
Filing date: 2022-01-24
Publication date: 2022-06-14

Abstract

The invention relates to a power transmission line path planning technology. The invention provides a power transmission line path optimization method based on intelligent image recognition, aiming at solving the problems that the path optimization of a power transmission line is influenced by expert experience or has low efficiency, and the technical scheme can be summarized as follows: firstly, loading a high-definition remote sensing satellite image, a digital elevation model and vector thematic data by adopting a GIS geographic information system; determining a starting point and a stopping point of the line, connecting the starting point and the stopping point to form an aerial line, and determining line parameters to obtain a design plan of the power transmission line; then intelligently identifying the surface feature elements by adopting a pre-established improved U-Net model; then, carrying out grid division, calculating, identifying and mapping out a project cost adjustment coefficient of the unit grid; and finally, calculating the cost of any side line or diagonal line of the unit grid, and equivalently solving the path with the minimum cost of the two-point line engineering by utilizing an A-x algorithm. The method has the advantages of being simple to use and suitable for power transmission line path planning.

Description

Power transmission line path optimization method based on intelligent image recognition

Technical Field

The invention relates to a power transmission line path planning technology, in particular to a power transmission line path optimization technology based on intelligent image recognition.

Background

In the traditional power transmission line path selection process, designers need to combine a topographic map and existing thematic data (meteorological conditions, planning regions, mining regions, established power lines and the like) to carry out indoor line selection. Because the information carried by the topographic map cannot completely reflect the actual distribution situation of the on-site ground objects, the designer needs to perform path optimization after further on-site exploration. In general, a plurality of feasible path schemes exist, designers carry out technical and economic comparison on the plurality of path schemes based on own experience, and make decisions after comprehensively considering various factors. However, due to the experience of each designer, the path schemes selected by different designers may be different, and the path scheme selected last may not be the optimal scheme.

In recent years, the path optimization of the power transmission line by adopting an expert scoring method and a Dijkstra algorithm is a feasible optimization method, and the problems existing in the traditional power transmission line path selection are partially solved. However, the following problems still remain: the expert scoring method is influenced by expert experience, and the scoring weight of each restriction factor cannot completely and accurately reflect the accurate influence on the quality of the path scheme; the Dijgres algorithm has too large search range and consumes longer time; all information contained in the remote sensing is influenced by manual plotting, and the efficiency is low.

Disclosure of Invention

The invention aims to solve the problems that the path optimization of the existing power transmission line is influenced by expert experience or has low efficiency, and provides a power transmission line path optimization method based on intelligent image recognition.

The technical scheme adopted by the invention for solving the technical problems is that the power transmission line path optimization method based on intelligent image recognition comprises the following steps:

step 1, loading a high-definition remote sensing satellite image, a digital elevation model and vector thematic data by adopting a GIS (geographic Information system) geographic Information system;

step 2, determining a starting point and a stopping point of the line, connecting the starting point and the stopping point to form an air line, and determining line parameters to obtain a design plan of the power transmission line;

step 3, setting an image intelligent identification range by taking an aerial line of a line as a center to obtain a line selection area, intelligently identifying surface feature elements in the line selection area of the high-definition remote sensing satellite image by adopting a pre-established improved U-Net model, marking in a vector form, and storing according to different vector layers to obtain an image identification result; the improved U-Net model is characterized in that on the basis of the U-Net model, a standard convolutional layer in a compression channel of the improved U-Net model is replaced by a Deformable Convolution (Deformable Convolution), in an extension channel of the improved U-Net model, a prediction result is output after one deconvolution-splicing-Convolution operation of each convolutional layer is completed, and finally, joint prediction is carried out on all prediction results to obtain final classification prediction;

step 4, carrying out grid division on the high-definition remote sensing satellite image, combining an image recognition result, a digital elevation model and a power transmission line design plan, calculating the altitude, the terrain and landform, the wind speed, the ice area, the forest area, the automobile and manpower distance of any edge or diagonal line of the grid of the recognition unit, and mapping a corresponding engineering cost adjustment coefficient for the high-definition remote sensing satellite image;

and 5, calculating the cost of any edge or diagonal of the unit grid through the single-kilometer reference cost of the line, the project cost adjustment coefficient, the cross spanning cost and the channel cleaning cost, solving the principle of the shortest path between two points in the network by utilizing an A-x algorithm, and equivalently solving the path with the minimum project cost of the two points of the line to obtain the optimal path of the line.

Specifically, in order to specify the vector thematic data, in step 1, the vector thematic data includes collected planning area and/or mining area and/or ecological sensitive area and/or grid space data.

Further, to specify the line parameters, in step 2, the line parameters include a line voltage level, a loop number, a conductor type number, and a conductor splitting number.

Specifically, in order to specify the surface feature elements, in step 3, the surface feature elements include houses and/or rivers and/or reservoirs and/or lakes and/or roads and/or railways and/or forest trees.

Further, to specifically explain how to pre-establish the improved U-Net model, in step 3, the training method of the pre-established improved U-Net model is as follows:

301, acquiring a high-definition remote sensing satellite image, carrying out sample marking on ground feature elements to be extracted in the high-definition remote sensing satellite image according to needs, randomly cutting the image and the manufactured sample mark into preset sizes, setting a training set, a testing set and a verification set according to a certain proportion, and enhancing a sample through space geometric transformation operation to obtain a deep learning sample;

step 302, constructing an improved U-Net model, and pre-training initialization parameters of the improved U-Net model on an ImageNet classification data set to obtain a pre-trained model;

and 303, inputting the deep learning sample set into a model after pre-training for feature learning to obtain a prediction probability distribution map, measuring a loss value between a classification result and a real ground feature element by adopting a cross entropy function, adopting an Adma optimization algorithm to aim at reducing the loss value, and finishing training when the loss value is reduced to a given threshold range to obtain an optimal model serving as an improved U-Net model.

Specifically, in order to provide a feasible preset size, in step 301, the preset size is 576 × 576; to provide a preferred ratio, the ratio is 7:2:1 in step 301; to account for the warping operation, the spatial geometry transformation operation includes cropping and/or rotation in step 301.

Further, to explain specifically how to use the pre-established improved U-Net model to perform intelligent recognition on the surface feature elements in the line selection area of the high-definition remote sensing satellite image, label the surface feature elements in a vector form, and store the surface feature elements according to different vector layers to obtain an image recognition result, in step 3, the pre-established improved U-Net model is used to perform intelligent recognition on the surface feature elements in the line selection area of the high-definition remote sensing satellite image, label the surface feature elements in a vector form, and store the surface feature elements according to different vector layers to obtain an image recognition result, specifically: and inputting the images to be classified in the line selection area of the high-definition remote sensing satellite images into an improved U-Net model for classification, splicing and vector labeling the classification results, and storing according to different vector layers to obtain an image identification result.

Specifically, to refine step 4, step 4 specifically includes:

step 401, performing mesh division on the remote sensing image, and regarding any unit mesh, taking each side line and each diagonal line of the unit mesh as each possible path in the unit mesh;

step 402, automatically calculating the altitude of the vertex of each unit grid according to a digital altitude model, taking the average value of the altitude of the starting point of the possible path as the calculated altitude of the possible path for any possible path in any unit grid, and mapping the engineering cost adjustment coefficient of the altitude by looking up a table according to the calculated altitude;

step 403, according to the digital elevation model and preset rules, searching for any possible path in any unit grid from the starting point of the possible path by taking a first distance as a search radius and a second distance as a step length until the end point of the possible path, intelligently identifying surface features according to the maximum elevation difference value and images in a circle formed by the search radius, classifying terrains, obtaining a construction cost adjustment coefficient of a terrains classification corresponding to each terrains classification by looking up a table, and obtaining the construction cost adjustment coefficient of the terrains classification of the possible path in a form of obtaining a weighted average of the construction cost adjustment coefficients of the terrains classification if a plurality of different types of terrains classifications exist in the possible path;

step 404, according to meteorological special data recorded in the early stage, aiming at any possible path in any unit grid, automatically extracting the wind speed and the ice area value of the area where the possible path is located, mapping the wind speed and the engineering cost adjustment coefficient of the ice area by looking up a table, and if a plurality of wind speed and ice area values exist in the possible path, obtaining the wind speed and the engineering cost adjustment coefficient of the ice area of the path in the grid in a form of solving the weighted average value of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values;

step 405, according to the flight path distance and the preset bending coefficient, calculating the total path length of the whole project, which is the flight path distance × the bending coefficient, and because the line project generally arranges 1 material station every fixed distance along the path line, if the path length is less than 2 times of the fixed distance, the automobile distance is path length/2, and if the path length is greater than or equal to 2 times of the fixed distance, the automobile distance is the fixed distance;

step 406, identifying road network information along the lines by an intelligent image identification technology, vectorizing the road network, aiming at any possible path in any unit grid, if the length of the possible path of the section is less than the average span, dividing the length of the possible path of the section by the average span and rounding to obtain the equal span which is more than or equal to 1, equally dividing the possible path of the section by the equal span, wherein the obtained starting point and the equal span are tower sites, connecting the starting point and the equal span with the roads in the road network respectively in the shortest distance to obtain the manpower transportation straight line path of each tower site, simultaneously aiming at any tower site, identifying the slope of the surface features and the transportation path by the image intelligence, classifying the terrain to obtain the transportation terrain of the manpower transportation straight line path of the tower site, and respectively connecting the starting point and the equal span with the roads in the road network in the shortest distance to obtain the length of the manpower transportation straight line path of each tower site, the method comprises the steps of firstly, multiplying the length of a manual transportation straight line path of each tower point by a bending coefficient corresponding to the transportation terrain of the manual transportation straight line path of the corresponding tower point, and then calculating an average value to obtain a manual transportation distance; the bending coefficient corresponding to each transport terrain is obtained by looking up a table, and the engineering cost adjustment coefficient is mapped according to the automobile haul distance and the manpower haul distance;

step 407, aiming at any possible path in any unit grid, judging whether a forest area exists on the possible path through image identification, and looking up a table to map out a corresponding engineering cost adjustment coefficient.

Still further, to explain the terrain classification, in step 403, the terrain is classified as flat ground, or hills, or mountains, or muddiness, or river network, or desert; to provide a preferred parameter for the first distance and/or the second distance, the first distance is 125 meters and the second distance is 250 meters.

Specifically, in order to provide a calculation formula for obtaining the engineering cost adjustment coefficients of the terrain classifications of the possible path in the form of obtaining the weighted average of the engineering cost adjustment coefficients of each terrain classification, in step 403, the calculation formula for obtaining the engineering cost adjustment coefficients of the terrain classifications of the possible path in the form of obtaining the weighted average of the engineering cost adjustment coefficients of each terrain classification is:

wherein, b refers to the project cost adjusting coefficient of the terrain classification of the possible path of the section, b_nThe engineering cost adjustment coefficient of the terrain classification corresponding to the nth terrain type is indicated; l is_nThe possible path length corresponding to the nth terrain type is referred to; m refers to the number of terrain types within the possible path.

Further, in order to provide a calculation formula for obtaining the wind speed and the engineering cost adjustment coefficient of the ice area on the possible path in the form of obtaining the weighted average of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values, in step 404, the calculation formula for obtaining the wind speed and the engineering cost adjustment coefficient of the ice area on the possible path in the form of obtaining the weighted average of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values is:

wherein c is the project cost adjustment coefficient of the wind speed of the section of possible path, d is the project cost adjustment coefficient of the ice area of the section of possible path, c_nIs the engineering cost adjustment coefficient, d, corresponding to the nth wind speed value_nThe engineering cost adjustment coefficient corresponding to the nth ice region numerical value is indicated; u shape_nMeans the possible path length, V, corresponding to the nth wind speed value_nThe possible path length corresponding to the nth ice region value is referred to; j refers to the number of different wind speed values within the possible path and q refers to the number of different ice bank values within the possible path.

Specifically, in order to provide a preferred fixed distance and a parameter of the meandering coefficient, in step 405, the fixed distance is 25 km; the tortuosity factor is 1.05.

Further, in order to provide a specific calculation formula for obtaining the human transport distance by multiplying the length of the human transport straight-line path of each tower point by the bending coefficient corresponding to the transport terrain of the human transport straight-line path of the corresponding tower point, and then calculating an average value, in step 406, the specific calculation formula for obtaining the human transport distance by multiplying the length of the human transport straight-line path of each tower point by the bending coefficient corresponding to the transport terrain of the human transport straight-line path of the corresponding tower point, and then calculating an average value is adopted as follows:

wherein p is the human transport distance, k is the same number, w_nBending modulus, x, for transport terrain for a human transport straight path of n tower sites_nIs the length of the straight path of human transport at the n-tower site.

Specifically, in order to provide a method for determining whether a forest area exists on a path in a grid, in step 407, the method for determining whether a forest area exists on a path in the grid through image recognition for any possible path in any grid is as follows: judging whether a forest region exists on the possible path through forest vector data extracted through image recognition, judging whether the length of the forest region is larger than or equal to half of the length of the possible path, if so, considering that the forest region exists on the possible path, and otherwise, considering that no forest region exists on the possible path.

Further, to provide a method for calculating a cross-over cost, in step 5, the method for calculating the cross-over cost includes: aiming at any possible path in any unit grid, different types of cross spanning are extracted through a vector image layer and an image recognition technology to obtain cross spanning types and corresponding quantity on the possible path, single spanning cost corresponding to each type needing cross spanning is obtained through table lookup, and then total cross spanning cost on the possible path is calculated according to the obtained cross spanning types and the corresponding quantity.

Still further, in order to provide a method for calculating a channel cleaning cost, in step 5, the method for calculating the channel cleaning cost includes: setting a safe horizontal distance aiming at any possible path in any unit grid, setting a removal projection range on a high-definition remote sensing satellite image according to the safe horizontal distance and the possible path, extracting houses in the removal projection range by adopting an image recognition technology, calculating removal cost according to the houses to obtain line house removal cost of the possible path, calculating land acquisition compensation and/or forest cutting cost according to the occupied area of a base tower on each tower site to obtain channel cleaning cost of each base tower, and finally adding the channel cleaning cost of each base tower and the line house removal cost of the possible path to obtain the channel cleaning cost of the possible path.

Specifically, in order to provide a specific calculation formula capable of calculating the cost of any one of the edges or diagonals of the unit grid through the road single-kilometer reference cost, the project cost adjustment coefficient, the cross-over cost and the channel cleaning cost, in step 5, the specific calculation formula capable of calculating the cost of any one of the possible paths of any one of the unit grids through the road single-kilometer reference cost, the project cost adjustment coefficient, the cross-over cost and the channel cleaning cost is as follows:

i-grid path cost is body cost plus other costs

Cost of the main body is equal to the path length multiplied by the cost of the main body

Cost of one kilometer of the body [ [ (a)_i+b_i+c_i+d_i+e_i+f_i)-5]X single kilometer investment reference price

Other cost is single kilometer other cost reference price multiplied by path length + channel cleaning cost + cross spanning cost

Wherein, the investment reference price of single kilometer body and the reference prices of other expenses of single kilometer are obtained by looking up the table of line parameters, a_i、b_i、c_i、d_i、e_i、f_iAnd respectively adjusting the engineering cost coefficients of the altitude, the terrain type, the wind speed, the ice area, the forest area and the automobile and manpower distance corresponding to the possible path in the i grid.

Further, to explain how to solve the principle of the shortest paths between two points in the network by using the a-algorithm, and equivalently solve the path with the minimum construction cost of the two-point line engineering, so as to obtain the path with the optimal line, in step 5, the principle of the shortest paths between two points in the network by using the a-algorithm is used, and equivalently solve the path with the minimum construction cost of the two-point line engineering, so as to obtain the path with the optimal line is as follows: any edge line or diagonal line in the unit grid is used as a possible path, and the manufacturing cost equivalence of the possible path is used for replacing the distance of each path in the A-x algorithm.

The invention has the beneficial effects that in the scheme of the invention, by adopting the power transmission line path optimization method based on intelligent image recognition, the remote sensing image is automatically recognized by adopting a deep learning technology, and aiming at the problems that the line selection area relates to various ground feature elements with different sizes and different scales, a U-Net model is taken as a basic model, and a deformable convolution sum FPN (characteristic pyramid network) is introduced, so that an improved U-Net model which can automatically adjust the detection position according to the shape and the size of a target and integrate multi-scale information is designed, the intelligent degree is high, and the working intensity of designers is reduced; the influence of each restriction factor on the engineering cost is quantized, the engineering cost of each path scheme can be accurately calculated, the precision is high, and the shortage of personal experience of designers is avoided; and the path with the minimum construction cost is solved by adopting an A-star algorithm, so that the search range is small and the efficiency is high.

Drawings

Fig. 1 is a flowchart of a power transmission line path optimization method based on intelligent image recognition according to the present invention.

Fig. 2 is a schematic diagram of calculating a diagonal distance from a node n to a terminal point in the embodiment of the present invention.

Fig. 3 is a schematic diagram of a next possible path of the intermediate node n in the embodiment of the present invention.

Detailed Description

The technical solution of the present invention is described in detail below with reference to the embodiments and the accompanying drawings.

The power transmission line path optimization method based on intelligent image recognition, disclosed by the invention, has a flow chart shown in figure 1, and comprises the following steps:

step 1, loading a high-definition remote sensing satellite image, a digital elevation model and vector thematic data by adopting a GIS geographic information system.

In order to specify the vector thematic data, in this step, the vector thematic data includes collected planning area and/or mining area and/or ecological sensitive area and/or grid space data and the like.

And 2, determining a line starting point and a line stopping point, connecting the starting point and the line stopping point to form an aerial line, and determining line parameters to obtain a power transmission line design plan.

To explain the line parameters specifically, in this step, the line parameters include the line voltage level, the number of loops, the conductor type, the number of conductor splits, and the like.

Step 3, setting an image intelligent identification range by taking an aerial line of a line as a center to obtain a line selection area, intelligently identifying surface feature elements in the line selection area of the high-definition remote sensing satellite image by adopting a pre-established improved U-Net model, marking in a vector form, and storing according to different vector layers to obtain an image identification result; the improved U-Net model is characterized in that on the basis of the U-Net model, a standard convolution layer in a compression channel of the U-Net model is replaced by a deformable convolution, in an extension channel of the U-Net model, a prediction result is output after one deconvolution-splicing-convolution operation is completed on each convolution layer, and finally, all prediction results are subjected to combined prediction to obtain final classification prediction.

To specify the land feature, the land feature in this step may include a house and/or a river and/or a reservoir and/or a lake and/or a road and/or a railway and/or vegetation, etc.

To specifically illustrate how to pre-establish the improved U-Net model, in step 3, the training method of the pre-established improved U-Net model may be:

and 303, inputting the deep learning sample set into a model after pre-training for feature learning to obtain a prediction probability distribution map, measuring a loss value between a classification result and a real ground feature element by adopting a cross entropy function, taking reduction of the loss value as a target by adopting an Adma optimization algorithm, and finishing training when the loss value is reduced to a given threshold value range to obtain an optimal model as an improved U-Net model.

Here, to provide a feasible preset size, in step 301, the preset size is preferably 576 × 576; to provide a preferred ratio, the ratio is preferably 7:2:1 in step 301; to explain the spatial geometric transformation operation, the spatial geometric transformation operation may include cropping and/or rotation, etc. in step 301.

In addition, in step 3, all the prediction results mentioned generally refer to 2-4 prediction results, and preferably 4 prediction results, because the structure of the model is the fusion of deconvolution and features going through 4 times upwards in the deconvolution process, so it is more appropriate to perform 4 deconvolution-concatenation-convolution operations to obtain 4 prediction results, and thus there are features from low level to high level. In the deep learning, after a prediction result is obtained each time, the model parameters are adjusted by back propagation, and then the adjusted model parameters are used to obtain a prediction result, and then the model parameters are adjusted … … by back propagation until the optimal model parameters are obtained, so as to obtain the optimal prediction result, so that in the model introduced with the FPN, all the prediction results are weighted each time in the model parameter adjustment process, then the loss function is obtained by comparing with the real result, then the model parameters are adjusted by back propagation, then all the next prediction results are obtained by weighting, the loss function is obtained by comparing with the real result, then the model parameters are adjusted … … until the loss function obtained at the last time does not drop, and then the model parameters are optimal (namely step 303).

Specifically, how to intelligently identify the feature elements in the line selection area of the high-definition remote sensing satellite image by using the pre-established improved U-Net model, label the feature elements in the vector form, and store the feature elements according to different vector layers to obtain an image identification result, then in step 3, intelligently identify the feature elements in the line selection area of the high-definition remote sensing satellite image by using the pre-established improved U-Net model, label the feature elements in the vector form, and store the feature elements according to different vector layers to obtain the image identification result may specifically be: and inputting the images to be classified in the line selection area of the high-definition remote sensing satellite images into an improved U-Net model for classification, splicing and vector labeling the classification results, and storing according to different vector layers to obtain an image identification result. Here, the vector notation means: because the classification result is output as a grid, the grid needs to be converted into a vector, and then attribute labeling is given.

It can be seen that in the present invention, the U-Net model is used as a basic network model, and the U-Net model is composed of a compression channel and an expansion channel, wherein the compression channel is a typical convolutional neural network structure, and is composed of a convolutional layer and a maximized pool layer, and is used for extracting image features layer by layer, the expansion channel gradually restores the details and position information of the image, generally firstly, a deconvolution operation is performed, then, the corresponding compressed feature maps are spliced to form a feature map with a size of 2 times, then, the convolutional layer is used for extracting features, the process is repeated for 4 times, then, the features are mapped into a two-dimensional output result in the last convolutional layer, but because the high-definition remote sensing satellite image comprises various ground feature elements such as houses and/or rivers and/or reservoirs and/or lakes and/or roads and/or railways and/or vegetation, the standard convolutional layer operation in the U-Net model is usually detected at a fixed position of the image feature map, and because the sizes of the targets corresponding to different positions are different, the standard convolutional layer operation in the U-Net model is not reasonable for the feature map with a complex target. Therefore, aiming at the problem, the deformable convolution is introduced on the basis of the U-Net model, the deformable convolution learns from the driving data of the task target, and the size and the position of the receptive field can be automatically adjusted according to the shape, the position and the size of the target by increasing the sampling offset (namely, an offset is added to the sampling position of an original feature map so as to change the position and the area of spatial sampling), so that the self-adaptive extraction capability of various ground feature elements in a high-definition remote sensing satellite image can be enhanced, namely, the identification and positioning capability is improved, and the improvement is made on a feature map layer; secondly, the scales of various surface feature elements are different, although the U-Net model utilizes some information from the previous layer of the coding stage, the generalization capability of the U-Net model to multi-scale information is very limited, so the invention uses the thought of a characteristic pyramid network for reference, has the thought of a plurality of prediction outputs, performs one prediction output on each layer of the coding stage, obtains a final loss function by weighting the inputs, and can utilize more scale information in the processes of back propagation and parameter updating, thereby further improving the classification precision.

And 4, carrying out grid division on the remote sensing image, combining an image recognition result, a digital elevation model and a power transmission line design plan, calculating the altitude, the terrain and landform, the wind speed, the ice area, the forest area and the automobile and manpower distance of any edge or diagonal line of the grid of the recognition unit, and mapping a corresponding engineering cost adjustment coefficient for the remote sensing image.

The step can be specifically as follows:

step 401, performing mesh division on the remote sensing image, and regarding any unit mesh, taking each edge line and each diagonal line of the unit mesh as each possible path in the unit mesh.

Step 402, automatically calculating the altitude of the vertex of each unit grid according to the digital altitude model, taking the average value of the altitude of the starting point of the possible path as the calculated altitude of the possible path for any possible path in any unit grid, and mapping the engineering cost adjustment coefficient of the altitude according to the calculated altitude by looking up a table.

Here, the table lookup refers to looking up a project cost adjustment coefficient table corresponding to the altitude, as shown in table 1.

TABLE 1 engineering cost adjustment coefficient table corresponding to altitude

Step 403, according to the digital elevation model and the preset rule, aiming at any possible path in any unit grid, searching from the starting point of the possible path by taking the first distance as the search radius and the second distance as the step length until the end point of the possible path, intelligently identifying the earth surface characteristics according to the maximum elevation difference value and the image in the circle formed by the search radius, classifying the terrain, obtaining the engineering cost adjustment coefficient of a terrain classification corresponding to each terrain classification by looking up a table, and obtaining the engineering cost adjustment coefficient of the terrain classification of the possible path in the form of obtaining the weighted average value of the engineering cost adjustment coefficients of each terrain classification if a plurality of different types of terrain classifications exist in the possible path.

Here, to explain the terrain classification, the terrain classification may be flat ground or hills or mountains or muddiness or river networks or deserts, etc.; to provide a preferred parameter for the first distance and/or the second distance, the first distance is preferably 125 meters and the second distance is preferably 250 meters.

In order to provide a formula for calculating the engineering cost adjustment coefficients of the terrain classifications of the possible route in the form of finding the weighted average of the engineering cost adjustment coefficients of the respective terrain classifications, in this step, the formula for calculating the engineering cost adjustment coefficients of the terrain classifications of the possible route in the form of finding the weighted average of the engineering cost adjustment coefficients of the respective terrain classifications may be:

wherein, b refers to the project cost adjusting coefficient of the terrain classification of the possible path of the section, b_nThe engineering cost adjustment coefficient of the terrain classification corresponding to the nth terrain type is indicated; l is a radical of an alcohol_nThe possible path length corresponding to the nth terrain type is referred to; m refers to the number of terrain types within the possible path.

In this step, the table lookup refers to the lookup of the engineering cost adjustment coefficient table corresponding to the terrain classification, as shown in table 2.

TABLE 2 landform classification corresponded engineering cost adjustment coefficient table

In addition, the preset rule referred in this step can be referred to a preset rule table, as shown in table 3.

TABLE 3 Preset rules Table

Numbering	Type of terrain	The unit meter is judged by the height difference delta h	Determined by the slope value theta, unit degree	Whether or not image recognition needs to be combined
					1	Flat ground	Δh<10	θ<3	Is that
2	Hills	10≤Δh<50	3≤θ<11	Whether or not
					3	Mountain land	50≤Δh<150	11≤θ<31	Whether or not
4	Alpine	150≤Δh<250	31≤θ<45	Whether or not
					5	Tsingling	250≤Δh	45≤θ	Whether or not
6	Mud and marsh	Δh<10	θ<3	Is that
					7	River net	Δh<10	θ<3	Is that
8	(Desert)	Δh<10	θ<3	Is that

And step 404, automatically extracting the wind speed and the ice area value of the area where the possible path is located aiming at any possible path in any unit grid according to meteorological special data input in the early stage, mapping the wind speed and the engineering cost adjustment coefficient of the ice area by looking up a table, and obtaining the wind speed and the engineering cost adjustment coefficient of the ice area of the path in the grid by adopting a form of solving a weighted average value of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values if a plurality of wind speed and ice area values exist in the possible path.

In order to provide a calculation formula for obtaining the wind speed of the possible path and the engineering cost adjustment coefficient of the ice area in a form of obtaining a weighted average value of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values, in this step, the calculation formula for obtaining the wind speed of the possible path and the engineering cost adjustment coefficient of the ice area in a form of obtaining a weighted average value of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values may be:

wherein c is the construction cost of the wind speed of the possible path of the sectionD is the engineering cost adjusting coefficient of the ice area of the possible path of the section, c_nIs the engineering cost adjustment coefficient, d, corresponding to the nth wind speed value_nThe engineering cost adjustment coefficient corresponding to the nth ice region numerical value is indicated; u shape_nMeans the possible path length, V, corresponding to the nth wind speed value_nThe possible path length corresponding to the nth ice region value is referred to; j refers to the number of different wind speed values within the possible path and q refers to the number of different ice bank values within the possible path.

In this step, the lookup table refers to the lookup of the engineering cost adjustment coefficient table corresponding to the wind speed value and the engineering cost adjustment coefficient table corresponding to the ice bank value, which are shown in tables 4 and 5, respectively.

Table 4 wind speed value corresponding engineering cost adjusting coefficient table

TABLE 5 engineering cost adjustment coefficient table corresponding to ice zone values

Step 405, according to the flight path distance and the preset bending coefficient, calculating the total path length of the whole project, which is the flight path distance × the bending coefficient, and because the line project generally arranges 1 material station every fixed distance along the path line, if the path length is less than 2 times of the fixed distance, the vehicle distance is the path length/2, and if the path length is greater than or equal to 2 times of the fixed distance, the vehicle distance is the fixed distance.

Here, to provide a preferred fixed distance and parameters of the meandering coefficient, the fixed distance may be 25 km; the tortuosity factor may be 1.05. In the case of a mountain line, the fixed distance may be set to 30km in the case of a poor traffic condition, and the fixed distance is preferably set to 25km in the general case. The tortuosity coefficient can also be adjusted according to the statistical data of the tortuosity coefficient of the line in the area.

Step 406, identifying road network information along the lines by an intelligent image identification technology, vectorizing the road network, aiming at any possible path in any unit grid, if the length of the possible path of the section is less than the average span, dividing the length of the possible path of the section by the average span and rounding to obtain the equal span which is more than or equal to 1, equally dividing the possible path of the section by the equal span, wherein the obtained starting point and the equal span are tower sites, connecting the starting point and the equal span with the roads in the road network respectively in the shortest distance to obtain the manpower transportation straight line path of each tower site, simultaneously aiming at any tower site, identifying the slope of the surface features and the transportation path by the image intelligence, classifying the terrain to obtain the transportation terrain of the manpower transportation straight line path of the tower site, and respectively connecting the starting point and the equal span with the roads in the road network in the shortest distance to obtain the length of the manpower transportation straight line path of each tower site, the method comprises the steps of firstly, multiplying the length of a manual transportation straight line path of each tower point by a bending coefficient corresponding to the transportation terrain of the manual transportation straight line path of the corresponding tower point, and then calculating an average value to obtain a manual transportation distance; the bending coefficient corresponding to each transport terrain is obtained by looking up a table, and the engineering cost adjustment coefficient is mapped according to the automobile transport distance and the manpower transport distance.

Here, the gradient of the transportation path may be calculated by obtaining a height difference between the tower location and the road in the road network through a digital elevation model, and calculating a horizontal distance between each tower location and the road in the road network through a GIS system, and the length of the linear path of the human transportation at each tower location obtained by connecting the start point and the bisector with the road in the road network at the shortest distance may be: if the slope of the transportation path is set as θ for any tower position (i.e. the starting point and the bisection point), the length of the straight line path of the manpower transportation at each tower position is equal to the horizontal distance/cos (θ) from the tower position to the road in the road network;

or the length of the manual transportation straight line path of each tower position point is equal to the square of the horizontal distance from the tower position point to the road in the road network and the square of the height difference between the tower position point and the road in the road network;

or, directly through GIS geographic information systemAnd comprehensively acquiring the slant distance between the two points so as to obtain the length of the manual transportation straight line path of each tower site. The method specifically comprises the following steps: when the straight-line distance between two points is calculated in a GIS (geographic information System), firstly, the longitude and latitude coordinates of the two points are converted into space rectangular coordinates, the coordinates of a certain tower position point are (x1, y1, z1), the coordinates of a point position connected with the corresponding shortest distance to a road are (x2, y2, z2), and then the coordinates of the tower position point are determined

The function of measuring the horizontal, vertical and slant distances between two points is the basic function in the GIS geographic information system.

In this step, the classification table for classifying the terrain can be referred to in table 3.

For providing a method, the length of the manpower transportation straight line path of each tower point is multiplied by the bending coefficient corresponding to the transportation terrain of the manpower transportation straight line path of the corresponding tower point, and then the average value is obtained to obtain a concrete calculation formula of the manpower distance, then the length of the manpower transportation straight line path of each tower point is multiplied by the bending coefficient corresponding to the transportation terrain of the manpower transportation straight line path of the corresponding tower point, and then the average value is obtained to obtain the concrete calculation formula of the manpower distance, which can be:

In this step, the bending coefficient table corresponding to the transport terrain is shown in table 6, and the engineering cost adjustment coefficient table corresponding to the automobile haul distance and the human haul distance is shown in table 7.

TABLE 6 transport topography corresponded bending modulus table

Numbering	Type of terrain	Coefficient of bending
			1	Flat ground	1.1
2	River network and mud	1.2
			3	Hills	1.3
4	Mountain land	1.5
			4	Mountain and grave mountain	1.8
5	(Desert)	1.5

TABLE 7 engineering cost adjustment coefficient table corresponding to automobile distance and manpower distance

In this step, since the tower positions are calculated in a rounded form, which results in a large error, it is recommended to set the length of the grid edge of the unit grid as an integer multiple of the average span as much as possible when dividing the unit grid, but there is a problem that there is no error in the grid edge but there is still an error in the diagonal, so the following method can be considered to reduce the error:

if the statistical average span value of a certain voltage class transmission line in a certain area is V, the side length of the grid is R, and the diagonal length is

Can pass through

And calculating the R value. From the R value, an optimal path can be calculated, and the average span V in the optimal path is the optimal path length/(node number-1). At this point, V is compared with V. Adjusting the cell grid side length by:

k₁+k₂＝2

k₁∈(0,2)，k₂∈(0,2)，n≥1

when n is a positive integer greater than or equal to 1, the tower positions are all on the top point of each unit grid (namely the starting point of a possible path), no other tower positions exist in the side line or the diagonal line of the unit grid, n is greater than or equal to 2, the top point of the unit grid is the tower position point, and other tower positions also exist in the side line or the diagonal line of the unit grid.

If V x>V, then can be adjusted to be large k₁Reducing R, dividing the grid by the R value again,and calculating an optimal path. If V<V, then k can be adjusted to be small₁Increasing R, and dividing the grid by using the R value again to calculate the optimal path.

Repeating the above process until the error requirement is met: abs (V x-V)/V < η% can be set. The eta value can be freely set according to specific conditions.

Here, to provide a method for determining whether a forest region exists on a path in a mesh, for any possible path in any mesh, the method for determining whether a forest region exists on the path in the mesh through image recognition may be: and judging whether a forest region exists on the possible path through forest vector data extracted by image recognition, judging whether the length of the forest region is more than or equal to half of the length of the possible path, if so, considering that the forest region exists on the possible path, and otherwise, considering that no forest region exists on the possible path.

In this step, the project cost adjustment coefficient table corresponding to the forest zone is shown in table 8.

TABLE 8 engineering cost adjustment coefficient table corresponding to forest zone

Voltage class	Forest-free area	Forest area
			110kv	1	1.230
220kV	1	1.210
			330kV	1	1.140
500kV	1	1.120

Here, to provide a method for calculating the cross-over cost, the method for calculating the cross-over cost may be: aiming at any possible path in any unit grid, different types of cross spans are extracted through a vector layer and an image recognition technology to obtain cross span types and corresponding quantities on the possible paths, single span cost corresponding to each type needing cross span is obtained through table lookup, and then total cross span cost on the possible paths is calculated according to the obtained cross span types and the corresponding quantities. When extracting the different types of cross spans, what technique is adopted may refer to the cross span type acquisition mode table, as shown in table 9. The table of the single-pass cost corresponding to the cross-over type is shown in table 10.

TABLE 9 Cross-spanning type acquisition mode Table

TABLE 10 Single stride expense TABLE corresponding to Cross stride types

In order to provide a method for calculating the channel cleaning cost, the method for calculating the channel cleaning cost may be: setting a safe horizontal distance aiming at any possible path in any unit grid, setting a removal projection range on a high-definition remote sensing satellite image along the possible path according to the safe horizontal distance (the possible path can be projected on a horizontal plane, the projection of the possible path is taken as a starting point, the projection perpendicular to the possible path on the horizontal plane is respectively increased to two sides of the projection of the possible path by a certain distance which is the set safe horizontal distance), extracting houses in the removal projection range by adopting an image recognition technology, calculating removal cost according to the houses, obtaining line house removal cost of the possible path, calculating land acquisition compensation and/or forest cutting cost according to the floor area of a base tower on each tower site, obtaining channel cleaning cost of each base tower, and finally adding the channel cleaning cost of each base tower and the line house removal cost of the possible path to obtain the channel cleaning cost of the possible path And (6) managing the cost. The cost unit prices of different channel cleaning types can be obtained by looking up a table, the cost unit price tables of different channel cleaning types are shown in table 11, the floor area of each base tower and the like can also be obtained by a channel cleaning cost calculation rule table, and the channel cleaning cost calculation rule table is shown in table 12.

TABLE 11 cost price table for different channel cleaning types

TABLE 12 calculation rule Table for channel cleaning cost

In order to provide a specific calculation formula for calculating the cost of any one edge or diagonal of a unit grid through the single-kilometer road cost, the project cost adjustment coefficient, the cross-over cost and the channel cleaning cost, in this step, the specific calculation formula for calculating the cost of any one possible path of any unit grid through the single-kilometer road cost, the project cost adjustment coefficient, the cross-over cost and the channel cleaning cost may be:

i-grid path cost is body cost plus other costs

The table of the body investment price and the other cost price references corresponding to the line parameters is shown in table 13, wherein the body investment price and the other cost price references are engineering values calculated based on the workload of the case that the line is on the flat ground, no ice is covered, the wind speed is 25m/s, the line does not pass through a forest area, the altitude is lower than 1500 meters, no cross span exists, no channel is cleared, the automobile distance is 5km, and no manpower distance exists.

Table 13 table of capital investment price and other standard price of fee corresponding to line parameter

Voltage class	Engineering scheme	Static investment (Wanyuan/km)	Capital investment price (ten thousand yuan/km)	Other cost basis (Wanyuan/km)
					500kv	4 x 400 sheet	162	116	46
500kv	4X 400 double	305	213	92
					500kv	4X 500 pairs	326	246	80
500kv	4 x 630 sheet	160	124	36
					500kv	4 x 630 bis	328	216	112
330kv	2 x 300 sheet	75	54	21
					330kv	2 x 300 pairs	136	86	50
330kv	2 x 400 sheet	78	55	23
					330kv	2 x 400 double	158	104	54
330kv	4 x 400 sheet	175	103	72
					330kv	4X 400 double	327	198	129
220kv	2 x 300 sheet	67	49	18
					220kv	2 x 300 pairs	116	86	30
220kv	2 x 400 sheet	73	51	22
					220kv	2 x 400 double	148	104	44
220kv	2 x 630 sheet	89	62	27
					220kv	2 x 630 bis	207	147	60
110kv	1 x 240 sheet	47	34	13
					110kv	1X 240 double	86	62	24
110kv	2 x 240 sheet	53	38	15
					110kv	2X 240 double	109	79	30
110kv	1 x 300 sheet	55	39	16
					110kv	1 x 300 pairs	95	68	27
110kv	2 x 300 pairs	120	89	31
					110kv	1 x 400 sheet	55	39	16
110kv	1X 400 pairs	98	68	30

To explain how to solve the principle of two-point shortest path in the network by using the a-algorithm, and equivalently solve the path with the minimum construction cost of two-point line engineering to obtain the optimal path of the line, in this step, the principle of two-point shortest path in the network by using the a-algorithm is used to solve the path with the minimum construction cost of two-point line engineering, and the obtained optimal path of the line means: any edge line or diagonal line in the unit grid is used as a possible path, and the manufacturing cost equivalence of the possible path is used for replacing the distance of each path in the A-x algorithm.

The A-algorithm is a typical heuristic search algorithm, is established on the basis of Dijkstra algorithm, and is widely applied to the algorithm in the field of path optimization. It uses heuristic information to decide which is the next node to expand and then searches for possible expansion outward along some edge segment that is considered most promising. Compared with Dijkstra algorithm, the method can greatly reduce the search range and reduce the calculation time.

When solving the problem of the shortest path between two points in the network, the a-x algorithm can be expressed by the following formula:

f*(n)＝g(n)+h*(n)

wherein f (n) is an evaluation function; g (n) is the shortest path value from the starting point to any node n, and h (n) is the shortest path estimate from n to the destination node.

h (n) is an important factor influencing the efficiency of the A algorithm, and h (n) cannot be too different from h (n), otherwise h (n) has no too strong distinguishing capability. A good heuristic strategy evaluation is: h (n) is less than h (n) and as close as possible to h (n).

The invention uses A-star algorithm to calculate the shortest path between two points in the network, and equivalently solves the path scheme with the minimum line cost. The distance of each path is equivalently replaced by the manufacturing cost of each possible path.

n to the minimum cost estimate of the target node:

h (n) ═ diagonal distance from node n to target node x single kilometer line reference price

The reference price of single-kilometer line is equal to the reference price of single-kilometer investment and the reference price of single-kilometer other expenses

And when the diagonal distance from the node n to the terminal is calculated, the no-passing areas are ignored, namely all areas can pass.

As shown in fig. 2, assuming that X < Y, the coordinates of the node n are (X1, Y1), the coordinates of the end point are (X2, Y2), and the side length of the cell grid is R, then:

X＝abs(X1-X2)

Y＝abs(Y1-Y2)

the next possible path for the intermediate node n is shown in fig. 3. The solid line path from the starting point to the node n is the determined path, and all possible paths of the node n in the next step are arrows from the node n in the graph to indicate paths. The selection principle of the paths in the unit grid is as follows: the path exclusion intersects with the exclusion passing areas (mining areas, planning areas, ecological sensitive area vector data); the included angle between the next step path of the node n and the previous grid path mn is more than or equal to 90 degrees; the step size is a grid of cells.

The algorithm implementation flow is as follows:

1) and putting the starting node into an open list (openlist), wherein the f value and the g value of the starting node are both 0.

2) The following procedure was repeated:

(1) and traversing the openlist, searching the node with the minimum f value, and taking the node as the current node to be processed.

(2) This node is added to the closed list (closelist) and removed from the openlist.

(3) Sequentially executing the following steps on each node adjacent to the current node:

a. if the neighbor is not passable or if the neighbor is already in the closed list, then nothing is performed and the next node is checked on.

b. If the adjacent node is not in the openlist, the node is added into the openlist, the parent node of the adjacent node is set as the current node (parent node: each node needs to record the position of the previous point reaching the point, the previous point is called as the parent node), and the g value and the f value of the adjacent node are saved.

c. If the adjacent node is in the openlist, judging whether the g value reaching the adjacent node through the current node is smaller than the original stored g value, if so, setting the father node of the adjacent node as the current node, and resetting the g value and the f value of the adjacent node.

(4) And (3) cycle end conditions: and when the end point is added into the openlist as the node to be checked, the shortest path is obtained and the search is stopped.

(5) Traversing from the terminal node along the father node, storing the coordinates of all nodes traversed, and connecting all the node coordinates to be the shortest path between the two points (at this time, because the cost of each possible path is equivalent to replace the distance of each path, the calculated path is the minimum cost path).

After the lowest cost path is calculated, the cost of the path scheme can be further corrected by fitting and adjusting coefficients alpha, beta and gamma so as to calculate more accurate line engineering cost value. The influence factors of the line cost are calculated without considering the influence of factors such as geological conditions, polluted areas, seismic intensity and the like, so that the manufacturing value is corrected by adopting the fitting adjustment coefficient beta.

When calculating the reference construction cost of the line, firstly, only selecting the typical construction cost of the Sichuan area as the reference to measure and calculate the adjustment coefficient (namely the tables), and not considering the typical construction cost characteristics of other areas in the country; and the market prices of materials, manpower and the like change constantly along with the change of time. Therefore, the line reference cost value is corrected by using the fitting adjustment coefficients α and γ.

The corrected line cost value is calculated as follows:

B'_i＝B^*+β(B_i-1)B^*＝B^*(1+β(B_i-1))

B_i＝(a_i+b_i+c_i+d_i+e_i+f_i)-5

B^*＝αrB^#

T'_i＝T^*+S_i

T^*＝γrT^#

wherein Z is the corrected line cost value, B^#The investment standard price of a single kilometer body is obtained, r is the path length in a unit grid, alpha, beta and gamma are fitting adjustment coefficients respectively, the alpha, the beta and the gamma are belonged to (0,1), and S is_iAdding a cross-over fee, T, to the channel clearing fee^#For a single kilometer of other cost reference prices, n is the number of unit grids passed by the line.

Claims

1. The power transmission line path optimization method based on intelligent image recognition is characterized by comprising the following steps of:

step 1, loading a high-definition remote sensing satellite image, a digital elevation model and vector thematic data by adopting a GIS (geographic information system);

step 3, setting an image intelligent identification range by taking an aerial line of a line as a center to obtain a line selection area, intelligently identifying surface feature elements in the line selection area of the high-definition remote sensing satellite image by adopting a pre-established improved U-Net model, marking in a vector form, and storing according to different vector layers to obtain an image identification result; the improved U-Net model is characterized in that on the basis of the U-Net model, a standard convolution layer in a compression channel of the improved U-Net model is replaced by a deformable convolution, in an extension channel of the improved U-Net model, a prediction result is output after one deconvolution-splicing-convolution operation of each convolution layer is completed, and finally, all prediction results are subjected to combined prediction to obtain final classification prediction;

2. The power transmission line path optimization method based on intelligent image recognition as claimed in claim 1, wherein in step 3, the training method of the pre-established improved U-Net model comprises:

3. The power transmission line path optimization method based on intelligent image recognition according to claim 1, wherein in step 3, the pre-established improved U-Net model is adopted to intelligently recognize surface feature elements in the line selection area of the high-definition remote sensing satellite image, the surface feature elements are labeled in a vector form and are stored according to different vector layers to obtain an image recognition result, and specifically, the method comprises the following steps: and inputting the images to be classified in the line selection area of the high-definition remote sensing satellite images into an improved U-Net model for classification, splicing and vector labeling the classification results, and storing according to different vector layers to obtain an image identification result.

4. The power transmission line path optimization method based on intelligent image recognition according to claim 1, wherein the step 4 specifically comprises:

step 404, according to meteorological special data entered in the early stage, aiming at any possible path in any unit grid, automatically extracting the wind speed and the ice area value of the area where the possible path is located, mapping the wind speed and the engineering cost adjustment coefficient of the ice area by looking up a table, and if a plurality of wind speed and ice area values exist in the possible path, obtaining the wind speed and the engineering cost adjustment coefficient of the ice area of the path in the grid in a mode of solving the weighted average value of the engineering cost adjustment coefficients corresponding to the wind speed and the ice area values;

step 405, calculating the total path length of the whole project as the flight line distance × the bending coefficient according to the flight line distance and the preset bending coefficient, and because the line project generally arranges 1 material station every fixed distance along the path line, if the path length is less than 2 times of the fixed distance, the vehicle distance is the path length/2, and if the path length is greater than or equal to 2 times of the fixed distance, the vehicle distance is the fixed distance;

5. The method for optimizing transmission line path based on intelligent image recognition as claimed in claim 4, wherein in step 403, the formula for calculating the project cost adjustment coefficients of the terrain classification of the possible path in the form of finding the weighted average of the project cost adjustment coefficients of each terrain classification is:

6. The power transmission line path optimization method based on intelligent image recognition as claimed in claim 4, wherein in step 404, the calculation formula for obtaining the project cost adjustment coefficients of the wind speed and the ice area of the possible path in the form of obtaining the weighted average of the project cost adjustment coefficients corresponding to the wind speed and the ice area numerical value is:

wherein c is the project cost adjustment coefficient of the wind speed of the section of possible path, d is the project cost adjustment coefficient of the ice area of the section of possible path, c_nRefers to the engineering cost adjustment coefficient, d, corresponding to the nth wind speed value_nThe engineering cost adjustment coefficient corresponding to the nth ice region value is indicated; u shape_nMeans the possible path length, V, corresponding to the nth wind speed value_nThe possible path length corresponding to the nth ice region value is referred to; j refers to the number of different wind speed values within the possible path and q refers to the number of different ice bank values within the possible path.

7. The power transmission line path optimization method based on intelligent image recognition of claim 4, wherein in step 406, the specific calculation formula for obtaining the human transport distance by first multiplying the length of the human transport straight-line path of each tower point by the bending coefficient corresponding to the transport terrain of the human transport straight-line path of the corresponding tower point and then calculating the average value is:

wherein p is the human transport distance, k is the same number, w_nBending modulus, x, for transport terrain for a human transport straight path of n tower sites_nIs n tower positionsThe length of the straight path of manual transport of the points.

8. The power transmission line path optimization method based on intelligent image recognition as claimed in claim 4, wherein in step 407, the method for determining whether a forest area exists on the path in any grid through image recognition for any possible path in any grid is as follows: judging whether a forest region exists on the possible path through forest vector data extracted through image recognition, judging whether the length of the forest region is larger than or equal to half of the length of the possible path, if so, considering that the forest region exists on the possible path, and otherwise, considering that no forest region exists on the possible path.

9. The power transmission line path optimization method based on intelligent image recognition as claimed in claim 4, wherein in step 5, the calculation method of the cross-over cost is as follows: aiming at any possible path in any unit grid, different types of cross spans are extracted through a vector layer and an image recognition technology to obtain cross span types and corresponding quantities on the possible paths, single span cost corresponding to each type needing cross span is obtained through table lookup, and then total cross span cost on the possible paths is calculated according to the obtained cross span types and the corresponding quantities.

10. The power transmission line path optimization method based on intelligent image recognition as claimed in claim 4, wherein in step 5, the calculation method of the channel cleaning cost is as follows: setting a safe horizontal distance aiming at any possible path in any unit grid, setting a removal projection range on a high-definition remote sensing satellite image along the possible path according to the safe horizontal distance, extracting houses in the removal projection range by adopting an image recognition technology, calculating removal cost according to the houses to obtain line house removal cost of the possible path, calculating land acquisition compensation and/or forest cutting cost according to the floor area of a base tower on each tower site respectively to obtain channel cleaning cost of each base tower, and finally adding the channel cleaning cost of each base tower and the line house removal cost of the possible path to obtain the channel cleaning cost of the possible path.

11. The method for optimizing transmission line path based on intelligent image recognition according to claim 4, wherein in step 5, the specific calculation formula for calculating the cost of any possible path of any unit grid through the single-kilometer reference cost of the line, the project cost adjustment coefficient, the cross-over cost and the channel cleaning cost is as follows:

i-grid path cost is body cost plus other costs

12. The transmission line path optimization method based on intelligent image recognition according to any one of claims 1 to 11, wherein in step 5, the principle of solving the shortest paths between two points in the network by using an a-x algorithm is used, the path with the minimum construction cost of two points of line engineering is equivalently solved, and the path with the optimal line is obtained by: any edge line or diagonal line in the unit grid is used as a possible path, and the manufacturing cost equivalence of the possible path is used for replacing the distance of each path in the A-x algorithm.