Building and road monomer segmentation method, electronic equipment and storage medium

Through the combination of the depth multi-scale feature extraction module, multi-head attention mechanism module and multi-loss function, the problems of blurred boundaries, smooth and discontinuousness in monolithic segmentation of buildings and roads are solved, and high-precision monolithic segmentation is achieved Effect.

CN119091303AActive Publication Date: 2024-12-06HARBIN AEROSPACE STAR DATA SYST TECH CO LTD

Patent Information

Application Number
CN202411234274.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-12-06
Estimated Expiration
2044-09-04

AI Technical Summary

Technical Problem

The prior art has problems of insufficient mining of multi-scale feature, false detection and missed detection in the monolithic segmentation of buildings and roads, and features are easily lost or misaligned during the encoding and decoding process, resulting in low accuracy.

Method used

A method of monolithization segmentation of buildings and roads is designed, including depth multi-scale feature extraction module DE_MSCAN, multi-head attention mechanism module Mhead_BRN, deep and shallow multi-scale void space pyramid module DS_MASPP and multi-loss function, and by constructing features Hierarchical extraction module and decoding module to optimize the model to improve segmentation accuracy.

Benefits of technology

It realizes efficient and accurate monolithic segmentation of buildings and roads, improves the model's ability to segment different sizes and shapes, and improves the overall classification accuracy and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091303A_ABST
    Figure CN119091303A_ABST
Patent Text Reader

Abstract

The invention discloses a building and road monomer segmentation method, electronic equipment and a storage medium, and belongs to the technical field of building monomer identification. In order to solve the problem that building and road monomer segmentation boundaries are fuzzy, the method comprises the following steps: acquiring high-resolution 1 # 2-meter multispectral image data of buildings and roads, and manually making a building and road monomer label data set; constructing a depth multi-scale feature extraction module; constructing a multi-head attention mechanism module of a building road; establishing a feature hierarchical extraction module; constructing a deep and shallow multi-scale cavity space pyramid module, and establishing a feature decoding module; designing a multi-loss function for guiding the optimization of the building-road singulation segmentation model; and constructing a building road singulation segmentation model DSMALnet, and carrying out model training to obtain an optimal building road singulation segmentation model. The method provided by the invention efficiently and accurately realizes the monomer segmentation of the building and the road in the multi-temporal satellite remote sensing image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of building monomer recognition, and in particular relates to a building and road monomer segmentation method, electronic equipment and storage medium. Background Art

[0002] As the urbanization process continues to accelerate, the construction land has also changed dramatically. Timely and accurate monitoring of the changes in construction land can accurately grasp the status of urban development. The number, area, location and ownership of rural houses are important basic data for applications such as rural land use planning, land use change monitoring, new rural construction and hollow village problem rectification. At present, the traditional methods of extracting construction land information using remote sensing images are mainly divided into three categories: the first category is to extract construction land information based on the spectral characteristics of typical objects or the relevant knowledge of spectral relationships, etc., using supervised classification, logical discrimination or decision tree methods. Although this type of method can remove bare land and extract construction land information, the efficiency of construction land information extraction and the universality of the model are restricted due to the large differences in spectral relationship characteristics between different regions or different images; the second category is to extract construction land information based on classification technology. Due to the spectral heterogeneity of construction land, the extraction accuracy of this method is often not high, and subsequent processing is required to improve the accuracy; the third category is to automatically extract construction land information by establishing a remote sensing index model. This method is not accurate for the extraction of building road monomers.

[0003] With the development of artificial intelligence, higher requirements are put forward for high-precision, high-efficiency and automatic individual extraction of buildings and roads. With the development of deep learning technology, traditional building-level road segmentation methods are gradually fading out of people's view. Due to the ability of convolutional neural networks in contextual information mining and multi-scale feature learning, it has become the standard method for segmenting buildings and roads. However, the widely used encoder-decoder neural network faces two technical obstacles. The first problem is that the multi-scale feature mining is insufficient, which is prone to false detection and missed detection. The second problem is that there will be loss and misalignment of features during the encoding and decoding process. Low-level features will be weakened during the extraction process, and features of different scales will also be misaligned. Summary of the invention

[0004] The problem to be solved by the present invention is that the boundaries of buildings and roads are blurred, too smooth and discontinuous, and the precision is low. A method for building and road segmentation, an electronic device, and a storage medium are proposed.

[0005] To achieve the above object, the present invention is implemented through the following technical solutions:

[0006] A method for segmenting buildings and roads into monomers, comprising the following steps:

[0007] S1. Collect 2-meter multispectral image data of buildings and roads from Gaofen-1, and manually create a dataset of individual building and road labels;

[0008] S2. Construct deep multi-scale feature extraction module DE_MSCAN;

[0009] S3. Construct a multi-head attention mechanism module Mhead_BRN for building roads;

[0010] S4. Based on the deep multi-scale feature extraction module DE_MSCAN constructed in step S2 and the multi-head attention mechanism module Mhead_BRN for building roads constructed in step S3, a feature hierarchical extraction module encoder is constructed;

[0011] S5. Construct a deep and shallow multi-scale void space pyramid module DS_MASPP, use LightHamHead and MLP methods for decoding, and build a feature decoding module decoder;

[0012] S6. Design multiple loss functions to guide the optimization of the building-road segmentation model;

[0013] S7. Based on the feature hierarchical extraction module encoder formed in step S4, the feature decoding module decoder formed in step S5 and the multi-loss function designed in step S6, a building road monomer segmentation model DSMALnet is constructed;

[0014] S8. Input the building and road monomer label data set produced in step S1 into the building road monomer segmentation model constructed in step S7 for model training to obtain the optimal building road monomer segmentation model;

[0015] S9. Use the optimal building and road monomer segmentation model obtained in step S8 to segment the buildings and roads in the target area.

[0016] Furthermore, the specific implementation method of step S2 includes the following steps:

[0017] S2.1. Set the DE_MSCAN module to consist of two batch normalization modules BN, one multi-head attention mechanism module Mhead_BRN for buildings and roads, and one framework flexible network module FFN. The calculation process expression is:

[0018] s DE_MSCAN =Mhead_BRN(BN(input t ))+input t (1)

[0019] x DE_MSCAN =FFN(BN(sDE_MSCAN ))+s DE_MSCAN (2)

[0020] Among them, input t Represents input feature information, s DE_MSCAN represents the feature information after BN and Mhead_BRN operations and matrix addition of the input feature information, x DE_MSCAN It means that after BN and FFN operations, it is combined with s DE_MSCAN Feature information for matrix addition, BN(.) represents batch normalization operation, Mhead_BRN(.) represents attention mechanism module calculation, FFN(.) represents FFN module calculation;

[0021] S2.2.FFN module calculation process expression is:

[0022] FFN=Conv 1×1 (GELU(Dw_Conv 3×3 (Conv 1×1 (s DE_MSCAN )))) (3)

[0023] Among them, Conv 1×1 (.) indicates a 1×1 convolution operation, GELU(.) indicates an activation function operation, and Dw_Conv 3×3 (.) represents a 3×3 depthwise convolution operation.

[0024] Furthermore, the specific implementation method of step S3 includes the following steps:

[0025] S3.1. The Mhead_BRN module is composed of two 1×1 convolutions, one GELU activation function, and one multi-attention mechanism Mhead_BR module. The calculation process expression of the Mhead_BRN module is:

[0026] S Mhead_BRN =GELU(Conv 1×1 (input h )) (4)

[0027] x Mhead_BRN =Conv 1×1 (Mhead_BR(S Mhead_BRN )) (5)

[0028] Among them, s Mhead_BRN Represents the feature information after 1×1 convolution and GELU activation function operation, x Mhead_BRN Represents the feature information after the Mhead_BR module and the 1×1 convolution operation, Conv 1×1(.) represents a 1×1 convolution operation, GELU(.) represents an activation function operation, and Mhead_BR(.) represents an Mhead_BR module calculation.

[0029] S3.2. Setting the Mhead_BR module consists of three parts, the specific structure is as follows:

[0030] S3.2.1. The first part is set as a multi-scale spatial and channel attention mechanism, which consists of 7 deep convolutions of different scales. The expression of the calculation process is:

[0031] s dw5 =Dw_Conv 5×5 (s Mhead_BRN ) (6)

[0032]

[0033] Among them, s dw5 、s dw7 、s dw11 、s dw21 Respectively represent the feature information of deep convolution with 5×5, 7×7, 11×11, and 21×21 convolution kernels, Dw_Conv i×j represents the depthwise convolution operation of the i×j convolution kernel, i,j∈[1,5,7,11,21];

[0034] S3.2.2. Set the second part as a multi-scale position attention mechanism, which consists of 2 average pooling, 2 maximum pooling, 4 normalized attention modules NAM, and 1 1×1 convolution. The expression of the calculation process is:

[0035]

[0036]

[0037] H_NAM,W_NAM=split(Conv 1×1 (concat(s h_d ,s w_d T ))) (10)

[0038] x pos =H_NAM×W_NAM (11)

[0039] Among them, s h_avgn ,s h_maxn ,S w_avgn ,S w_maxn They represent the features obtained by performing average pooling in the height direction, maximum pooling in the height direction, average pooling in the width direction, and maximum pooling in the width direction and then performing NAM calculation, respectively.h_d ,S w_d They represent the features obtained by performing matrix dot multiplication on the features in the height and width directions obtained by formula 8, H_NAM and W_NAM represent the features obtained by formula 9 after matrix connection, 1×1 convolution, and matrix segmentation along the height and width, respectively. pos represents the weight matrix after matrix multiplication of the features obtained by formula 10, H_AvgPool(.) and W_AvgPool(.) represent the average pooling operations in the height and width directions respectively, H_MaxPool(.) and W_MaxPool(.) represent the maximum pooling operations in the height and width directions respectively, NAM(.) represents the NAM module calculation, Conv 1×1 (.) represents a 1×1 convolution operation, split(.) represents a matrix split operation, T represents a matrix transpose operation, and ⊙ represents a matrix dot multiplication calculation;

[0040] The NAM calculation formula is as follows:

[0041] Y(BN)=ω λ (BN(Y(F))) (12)

[0042]

[0043] Among them, Y(F) represents the input feature map, λ represents the scale factor of each channel, ω λ Represents the obtained weight, μ B and σ B Represent the mean and standard deviation of the mini-batch, r and β represent the trainable finite transformation scale and offset, BN(s) represents the scale factor, ∈ represents the penalty factor, Bin out Represents the output features of the previous layer;

[0044] S3.2.3. Add the four weight matrices obtained in the first part to the one weight matrix obtained in the second part, and then perform a 1×1 convolution operation. The obtained weight matrix is ​​matrix-multiplied with the input feature information. The expression of the calculation process is:

[0045] x dw =Conv 1×1 (s dw5 +s dw7 +s dw11 +s dw21 +x pos )×s Mhead_BRN (15)

[0046] Among them, x dw Represents the feature information operated by the Mhead_BRN module.

[0047] Furthermore, the specific implementation method of step S4 is to perform a convolution operation with a 3×3 convolution kernel and a step size of 2 on the input data set to obtain the calculation result Stage0 of the 0th stage; Stage0 is calculated for 4 stages, that is, the DE_MSCAN operation designed in step S2 is performed 2 times, 2 times, 4 times, and 2 times, and after each stage, a convolution operation with a 3×3 convolution kernel and a step size of 2 is connected for downsampling, and the calculation results of the 4 stages are respectively the calculation result Stage1 of the 1st stage, the calculation result Stage2 of the 2nd stage, the calculation result Stage3 of the 3rd stage, and the calculation result Stage4 of the 4th stage.

[0048] Furthermore, the DS_MASPP module of step S5 is divided into two parts, and the specific implementation method is as follows:

[0049] S5.1. Set the first part of the DS_MASPP module to shallow multi-scale information fusion, which consists of a 3×3 convolution kernel, a batch normalization BN, and a Relu6 activation function. The expression of the calculation process is:

[0050] Stage0_1=Relu6(BN(Conv 3×3 (concat(Stage0,bicubic(Stage1))))) (16)

[0051] Among them, Stage0_1 represents the shallow multi-scale information fusion feature, bicubic(.) represents the bicubic interpolation operation, concat(.,.) represents the matrix connection operation, Conv 3×3 (.) represents a 3×3 convolution operation, BN(.) represents a batch normalization operation, and Relu6(.) represents a Relu6 activation function operation;

[0052] S5.2. The second part of the DS_MASPP module is set to mid-deep multi-scale information fusion, which consists of 3 3×3 dilated convolutions with separation rates of 3, 5, and 7, 4 batch normalization BNs, 4 relu6 activation functions, and 2 3×3 convolutions. The expression of the calculation process is:

[0053] Stage2_4 = Relu6(BN(Conv 3×3 concat(Stage2,bicubic(Stage3, Stage4)))) (17)

[0054]

[0055] Stage d =Conv 3×3 (concat(srate3 ,s rate5 ,s rate7 )) (19)

[0056] Among them, Stage2_4 represents the features after the middle and deep features are connected, s rate3 ,s rate5 ,s rate7 Respectively represent the feature information after the dilated convolution operation with separation rates of 3, 5, and 7, Stage d Denotes the features after fusion of mid- and deep-level multi-scale information, D_Conv rate=3 、D_Conv rate=5 、D_Conv rate=7 They represent dilated convolution operations with separation rates of 3, 5, and 7, respectively;

[0057] S5.3. Combine Stage0_1 obtained in step S5.1 with Stage d Stored in the List list, and then input into LightHamHead and MLP for feature decoding to obtain the predicted building road result y out .

[0058] Furthermore, the specific implementation method of step S6 is to convert the artificial building and road monomer label data y obtained in step S1 into label and y obtained in step S5 out Multi-classification cross entropy loss Calculation, Dice loss function Calculation, edge detection loss function After calculation, the losses are added to obtain multiple loss functions To guide model optimization, the calculation expression is:

[0059]

[0060] Multi-classification cross entropy loss The calculation formula is as follows:

[0061]

[0062] Where M represents the number of categories; y o,c represents a binary indicator that is 1 if category c is the correct category for the object and 0 otherwise; p o,c Indicates the probability that the model predicts that the object belongs to category c;

[0063] Dice loss function The calculation formula is as follows:

[0064]

[0065] Among them, F p represents the predicted label set; F gt represents the true label set;

[0066] Edge Detection Loss Function The calculation formula is as follows:

[0067]

[0068] Among them, using the Laplace operator Label y label and the predicted y out For edge extraction, the background is 0 and the edge is 1. The calculation formula is as follows:

[0069]

[0070] Among them, f(k,u) is a pixel point in the image, and f(k-1,u), f(k+1,u), f(k,u-1), and f(k,u+1) are the four adjacent pixel points around f(k,u).

[0071] Furthermore, step S7 constructs the building road monomer segmentation model DSMALnet, which is divided into two stages: encoder and decoder; the encoder is composed of 5 3×3 convolutions for downsampling and 10 DE_MSCANs. After the first four downsamplings, 2, 2, 4, and 2 DE_MSCAN calculations are performed respectively to realize multi-scale, multi-space, and multi-position feature information extraction; the decoder inputs each downsampled feature information into the DS_MASPP module, and then inputs it into the lightweight Hamburger and multi-layer perceptron MLP for feature decoding to obtain the prediction result; the loss function uses multiple loss functions to evaluate the model loss.

[0072] Furthermore, in step S8, the building and road monomer label data set produced in step S1 is input into the building road monomer segmentation model DSMALnet constructed in step S7 for model training to obtain the optimal building road monomer segmentation model; the average intersection over union ratio mIoU, category average pixel accuracy mPA, overall classification accuracy Accuracy, accuracy Precision, recall rate Recall, and F1 score are used to evaluate the accuracy of the model segmentation.

[0073] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for segmenting buildings and roads into individual units when executing the computer program.

[0074] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for segmenting buildings and roads into monomers.

[0075] Beneficial effects of the present invention:

[0076] The invention discloses a method for the individual segmentation of buildings and roads. The invention designs a deep multi-scale feature extraction module DE_MSCAN, realizes the multi-level and multi-scale feature information extraction, and more effective correlation between feature information and target information; designs a multi-head attention mechanism module Mhead_BRN for buildings and roads, which is used to obtain and share context information in multiple dimensions and multiple scales, effectively retain high-quality information, and more accurately guide the model feature information collection; designs a deep and shallow multi-scale void space pyramid module DS_MASPP, which is used to improve the model's segmentation ability for features of different sizes and shapes; designs multiple loss functions to guide the model to perform high-precision and high-efficiency optimization. The method of the invention can efficiently and accurately realize the individual segmentation of buildings and roads in multi-phase satellite remote sensing images, and provide certain technical support for many fields such as urban planning, rural improvement, non-agricultural and non-grain, and ecological and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a flow chart of a method for segmenting buildings and roads into individual units according to the present invention;

[0078] Figure 2 This is the architecture diagram of the building road monomer segmentation model DSMALnet of the present invention;

[0079] Figure 3 This is the architecture diagram of the deep multi-scale feature extraction module DE_MSCAN of the present invention;

[0080] Figure 4 This is the architecture diagram of the multi-head attention mechanism module Mhead_BRN for building roads described in the present invention;

[0081] Figure 5 This is the architecture diagram of the deep and shallow multi-scale void space pyramid module DS_MASPP of the present invention;

[0082] Figure 6 These are the accuracy evaluation graphs of the optimal building road monomer segmentation model described in the present invention, where (a) is the mIoU evaluation graph, (b) is the mPA evaluation graph, (c) is the Accuracy evaluation graph, (d) is the Precision evaluation graph, (e) is the Recall evaluation graph, and (f) is the F1 score evaluation graph;

[0083] Figure 7This is a comparison diagram of the building road monomer segmentation model results described in the present invention and the current mainstream model results. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0085] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0086] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 -Attached Figure 7 The detailed instructions are as follows:

[0087] Embodiment 1:

[0088] A method for segmenting buildings and roads into monomers, comprising the following steps:

[0089] S1. Collect 2-meter multispectral image data of buildings and roads from Gaofen-1, and manually create a dataset of individual building and road labels;

[0090] Furthermore, in step S1, based on the Gaofen-1 2-meter multispectral image data of the main urban area of ​​Harbin, ArcGIS10.7 is used to perform individual vector annotation of buildings and roads, and then the vector-to-raster function is used to convert the annotated vector data into tiff raster data as label data, where 0 represents background, 1 represents road, and 2 represents building. The Gaofen-1 2-meter multispectral image data and the produced label data are cut into 512×512 sizes, and 70% is used as a training data set and 30% as a test data set, and finally the production of individual label data sets of buildings and roads is completed;

[0091] S2. Construct deep multi-scale feature extraction module DE_MSCAN;

[0092] Furthermore, the specific implementation method of step S2 includes the following steps:

[0093] S2.1. Set the DE_MSCAN module to consist of two batch normalization modules BN, one multi-head attention mechanism module Mhead_BRN for buildings and roads, and one framework flexible network module FFN. The calculation process expression is:

[0094] s DE_MSCAN =Mhead_BRN(BN(input t ))+input t (1)

[0095] x DE_MSCAN =FFN(BN(s DE_MSCAN ))+s DE_MSCAN (2)

[0096] Among them, input t represents the input feature information, S DE_MSCAN represents the feature information after BN and Mhead_BRN operations and matrix addition of the input feature information, x DE_MSCAN It means that after BN and FFN operations, it is combined with s DE_MSCAN Feature information for matrix addition, BN(.) represents batch normalization operation, Mhead_BRN(.) represents attention mechanism module calculation, FFN(.) represents FFN module calculation;

[0097] S2.2.FFN module calculation process expression is:

[0098] FFN=Conv 1×1 (GELU(Dw_Conv 3×3 (Conv 1×1 (s DE_MACAN )))) (3)

[0099] Among them, Conv 1×1 (.) indicates a 1×1 convolution operation, GELU(.) indicates an activation function operation, and Dw_Conv 3×3 (.) represents a 3×3 depthwise convolution operation.

[0100] S3. Construct a multi-head attention mechanism module Mhead_BRN for building roads;

[0101] Furthermore, the specific implementation method of step S3 includes the following steps:

[0102] S3.1. The Mhead_BRN module is composed of two 1×1 convolutions, one GELU activation function, and one multi-attention mechanism Mhead_BR module. The calculation process expression of the Mhead_BRN module is:

[0103] s Mhead_BRN =GELU(Conv 1×1 (input h )) (4)

[0104] x Mhead_BRN =Conv 1×1 (Mhead_BR(s Mhead_BRN )) (5)

[0105] Among them, s Mhead_BRN Represents the feature information after 1×1 convolution and GELU activation function operation, x Mhead_BRN Represents the feature information after the Mhead_BR module and the 1×1 convolution operation, Conv 1×1 (.) indicates 1×1 convolution operation, GELU(.) indicates activation function operation, and Mhead_BR(.) indicates Mhead_BR module calculation;

[0106] S3.2. Setting the Mhead_BR module consists of three parts, the specific structure is as follows:

[0107] S3.2.1. The first part is set as a multi-scale spatial and channel attention mechanism, which consists of 7 deep convolutions of different scales. The expression of the calculation process is:

[0108] s dw5 =Dw_Conv 5×5 (s Mhead_BRN ) (6)

[0109]

[0110] Among them, s dw5 、s dw7 、s dw11 、s dw21 Respectively represent the feature information of deep convolution with 5×5, 7×7, 11×11, and 21×21 convolution kernels, Dw_Conv i×j represents the depthwise convolution operation of the i×j convolution kernel, i,j∈[1,5,7,11,21];

[0111] S3.2.2. Set the second part as a multi-scale position attention mechanism, which consists of 2 average pooling, 2 maximum pooling, 4 normalized attention modules NAM, and 1 1×1 convolution. The expression of the calculation process is:

[0112]

[0113] H_NAM,W_NAM=split(Conv 1×1 (concat(sh_d ,s w_d T ))) (10)

[0114] x pos =H_NAM×W_NAM (11)

[0115] Among them, s h_avgn ,s h_maxn ,s w_avgn ,s w_maxn They represent the features obtained by performing average pooling in the height direction, maximum pooling in the height direction, average pooling in the width direction, and maximum pooling in the width direction and then performing NAM calculation, respectively. h_d ,s w_d They represent the features obtained by performing matrix dot multiplication on the features in the height and width directions obtained by formula 8, H_NAM and W_NAM represent the features obtained by formula 9 after matrix connection, 1×1 convolution, and matrix segmentation along the height and width, respectively. pos represents the weight matrix after matrix multiplication of the features obtained by formula 10, H_AvgPool(.) and W_AvgPool(.) represent the average pooling operations in the height and width directions respectively, H_MaxPool(.) and W_MaxPool(.) represent the maximum pooling operations in the height and width directions respectively, NAM(.) represents the NAM module calculation, Conv 1×1 (.) represents a 1×1 convolution operation, split(.) represents a matrix split operation, T represents a matrix transpose operation, and ⊙ represents a matrix dot multiplication calculation;

[0116] The NAM calculation formula is as follows:

[0117] Y(BN)=ω λ (BN(Y(F))) (12)

[0118]

[0119] Among them, Y(F) represents the input feature map, λ represents the scale factor of each channel, ω λ Represents the obtained weight, μ B and σ B Represent the mean and standard deviation of the mini-batch, r and β represent the trainable finite transformation scale and offset, BN(s) represents the scale factor, ∈ represents the penalty factor, Bin out Represents the output features of the previous layer;

[0120] S3.2.3. Add the four weight matrices obtained in the first part to the one weight matrix obtained in the second part, and then perform a 1×1 convolution operation. The obtained weight matrix is ​​matrix-multiplied with the input feature information. The expression of the calculation process is:

[0121] x dw =Conv 1×1 (s dw5 +s dw7 +s dw11 +s dw21 +x pos )×s Mhead_BRN (15)

[0122] Among them, x dw Represents the feature information operated by the Mhead_BRN module.

[0123] S4. Based on the deep multi-scale feature extraction module DE_MSCAN constructed in step S2 and the multi-head attention mechanism module Mhead_BRN for building roads constructed in step S3, a feature hierarchical extraction module encoder is constructed;

[0124] Further, the specific implementation method of step S4 is to perform a convolution operation with a 3×3 convolution kernel and a step size of 2 on the input data set to obtain the calculation result Stage0 of the 0th stage; Stage0 is further calculated in 4 stages, that is, the DE_MSCAN operation designed in step S2 is performed 2 times, 2 times, 4 times, and 2 times, and after each stage, a convolution operation with a 3×3 convolution kernel and a step size of 2 is connected for downsampling, and the calculation results of the 4 stages are respectively the calculation result Stage1 of the first stage, the calculation result Stage2 of the second stage, the calculation result Stage3 of the third stage, and the calculation result Stage4 of the fourth stage;

[0125] S5. Construct a deep and shallow multi-scale void space pyramid module DS_MASPP, use LightHamHead and MLP methods for decoding, and build a feature decoding module decoder;

[0126] The DS_MASPP module of the further step S5 is divided into two parts, and the specific implementation method is as follows:

[0127] S5.1. Set the first part of the DS_MASPP module to shallow multi-scale information fusion, which consists of a 3×3 convolution kernel, a batch normalization BN, and a Relu6 activation function. The expression of the calculation process is:

[0128] Stage0_1=Relu6(BN(Conv 3×3(concat(Stage0,bicubic(Stage1))))) (16)

[0129] Among them, Stage0_1 represents the shallow multi-scale information fusion feature, bicubic(.) represents the bicubic interpolation operation, concat(.,.) represents the matrix connection operation, Conv 3×3 (.) represents a 3×3 convolution operation, BN(.) represents a batch normalization operation, and Relu6(.) represents a Relu6 activation function operation;

[0130] S5.2. The second part of the DS_MASPP module is set to mid-deep multi-scale information fusion, which consists of 3 3×3 dilated convolutions with separation rates of 3, 5, and 7, 4 batch normalization BNs, 4 relu6 activation functions, and 2 3×3 convolutions. The expression of the calculation process is:

[0131] Stage2_4 = Relu6(BN(Conv 3×3 concat(Stage2,bicubic(Stage3, Stage4)))) (17)

[0132]

[0133] Stage d =Conv 3×3 (concat(s rate3 ,s rate5 ,s rate7 )) (19)

[0134] Among them, Stage2_4 represents the features after the middle and deep features are connected, s rate3 ,s rate5 ,s rate7 Respectively represent the feature information after the dilated convolution operation with separation rates of 3, 5, and 7, Stage d Denotes the features after fusion of mid- and deep-level multi-scale information, D_Conv rate=3 、D_Conv rate=5 、D_Conv rate=7 They represent dilated convolution operations with separation rates of 3, 5, and 7, respectively;

[0135] S5.3. Combine Stage0_1 obtained in step S5.1 with Stage d Stored in the List list, and then input into LightHamHead and MLP for feature decoding to obtain the predicted building road result y out .

[0136] S6. Design multiple loss functions to guide the optimization of the building-road segmentation model;

[0137] Furthermore, the specific implementation method of step S6 is to convert the artificial building and road monomer label data y obtained in step S1 into label and y obtained in step S5 out Multi-classification cross entropy loss Calculation, Dice loss function Calculation, edge detection loss function After calculation, the losses are added to obtain multiple loss functions To guide model optimization, the calculation expression is:

[0138]

[0139] Multi-classification cross entropy loss The calculation formula is as follows:

[0140]

[0141] Where M represents the number of categories; y o,c represents a binary indicator that is 1 if category c is the correct category for the object and 0 otherwise; p o,c Indicates the probability that the model predicts that the object belongs to category c;

[0142] Dice loss function The calculation formula is as follows:

[0143]

[0144] Among them, F p represents the predicted label set; F gt represents the true label set;

[0145] Edge Detection Loss Function The calculation formula is as follows:

[0146]

[0147] Among them, using the Laplace operator Label y label and the predicted y out For edge extraction, the background is 0 and the edge is 1. The calculation formula is as follows:

[0148]

[0149] Among them, f(k,u) is a pixel point in the image, and f(k-1,u), f(k+1,u), f(k,u-1), and f(k,u+1) are the four adjacent pixel points around f(k,u).

[0150] S7. Based on the feature hierarchical extraction module encoder formed in step S4, the feature decoding module decoder formed in step S5 and the multi-loss function designed in step S6, a building road monomer segmentation model DSMALnet is constructed;

[0151] Furthermore, step S7 constructs the building road monomer segmentation model DSMALnet, which is divided into two stages: encoder and decoder; the encoder is composed of 5 3×3 convolutions for downsampling and 10 DE_MSCANs. After the first four downsamplings, 2, 2, 4, and 2 DE_MSCAN calculations are performed respectively to realize multi-scale, multi-space, and multi-position feature information extraction; the decoder inputs each downsampled feature information into the DS_MASPP module, and then inputs it into the lightweight Hamburger and multi-layer perceptron MLP for feature decoding to obtain the prediction result; the loss function uses multiple loss functions to evaluate the model loss.

[0152] S8. Input the building and road monomer label data set produced in step S1 into the building road monomer segmentation model constructed in step S7 for model training to obtain the optimal building road monomer segmentation model;

[0153] Furthermore, in step S8, the building and road monomer label data set produced in step S1 is input into the building road monomer segmentation model DSMALnet constructed in step S7 for model training to obtain the optimal building road monomer segmentation model; the average intersection over union ratio mIoU, category average pixel accuracy mPA, overall classification accuracy Accuracy, accuracy Precision, recall rate Recall, and F1 score are used to evaluate the accuracy of the model segmentation.

[0154] Furthermore, the specific calculation formula is as follows:

[0155]

[0156] Among them, k represents the number of categories, p ij Indicates that i is predicted as j, which is a false negative FN; p ji Indicates that j is predicted to be i, which is a false positive FP; p ii It means that i is predicted to be i, which is the true TP;

[0157]

[0158] Among them, k represents the number of categories, i represents the classification category, and CPA represents the category pixel accuracy;

[0159]

[0160] Among them, k represents the number of categories, i represents the actual classification category, j represents the predicted classification category, and p ij Indicates that i is predicted to be j, which is a false negative; p ii It means that i is predicted to be i, which is the true TP;

[0161]

[0162] Among them, TP represents the positive samples that are predicted to be correctly classified, i.e. true positives, and FP represents the negative samples that are predicted to be incorrectly classified, i.e. false positives;

[0163]

[0164] Among them, TP represents the positive samples that are predicted to be correctly classified, that is, true positive samples, and FN represents the positive samples that are predicted to be incorrectly classified, that is, false negative;

[0165]

[0166] Among them, Precision represents the accuracy and Recall represents the recall rate.

[0167] S9. Use the optimal building and road monomer segmentation model obtained in step S8 to segment the buildings and roads in the target area.

[0168] To facilitate understanding of the technical effects of the present invention, the comparative effects of the method of the present invention and the prior art are shown in the following table:

[0169] Table 1 Comparison of the effects of the methods of the embodiments of the invention and the prior art

[0170]

[0171] It can be seen from Table 1 that compared with other public methods, the model accuracy of the method proposed in the present invention has been greatly improved, and all evaluation indicators are optimal, among which the overall classification accuracy OA has reached 95.19%, the average intersection over Union ratio mIoU of building monomer segmentation has reached 73.12%, and the average intersection over Union ratio mIoU of road monomer segmentation has reached 77.87%; the average pixel accuracy mPA of building monomer segmentation has reached 85.56%, and the average pixel accuracy mPA of road monomer segmentation has reached 88.20%; the accuracy Precision of building monomer segmentation has reached 83.41%, and the accuracy Precision of road monomer segmentation has reached 86.93%;

[0172] The recall rate of building monomerization reached 85.56%, and the recall rate of road monomerization reached 88.20%; the F1 score of building monomerization reached 84.49%, and the F1 score of road monomerization reached 87.56%.

[0173] In order to understand the technical effect of each module of the present invention, an ablation experiment is used to conduct a comparative experiment between modules. The experimental results are shown in the following table:

[0174] Table 2 Comparison of ablation experiments of the methods of the embodiments of the invention

[0175]

[0176]

[0177] As can be seen from Table 2, the ablation experiment of the method proposed in the present invention is based on the proposed deep and shallow multi-scale void space pyramid module DS_MASPP, the multi-head attention mechanism module Mhead_BRN for buildings and roads, and the multi-loss function The experimental accuracy of both individual and pairwise combinations is improved compared to the base, and the accuracy based on the combination of all extracted modules is the highest, indicating that the model containing all modules has achieved the optimality and is helpful for the individual extraction of buildings and roads.

[0178] In summary: this embodiment makes full use of the design of the deep multi-scale feature extraction module DE_MSCAN to achieve multi-level and multi-scale feature information extraction, and more effective correlation between feature information and target information; designs the multi-head attention mechanism module Mhead_BRN for building roads, which is used to obtain and share context information in multiple dimensions and scales, effectively retain high-quality information, and more accurately guide the collection of model feature information; designs the deep and shallow multi-scale void space pyramid module DS_MASPP to improve the model's ability to segment features of different sizes and shapes; designs multiple loss functions To guide the model to optimize with high precision and efficiency. It solves the problems of blurred boundaries, oversmoothness, discontinuity and low precision of building and road segmentation.

[0179] Embodiment 2:

[0180] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for segmenting buildings and roads into individual units as described in Example 1 when executing the computer program.

[0181] The computer device of the present invention may be a device including a processor and a memory, such as a single chip microcomputer including a central processing unit, etc. Moreover, the processor is used to implement the steps of the above-mentioned method for segmenting buildings and roads into monomers when executing a computer program stored in the memory.

[0182] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0183] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0184] Embodiment 3:

[0185] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for segmenting buildings and roads into monomers as described in Example 1.

[0186] The computer-readable storage medium of the present invention can be any form of storage medium that can be read by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. A computer program is stored on the computer-readable storage medium. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned method for segmenting buildings and roads can be implemented.

[0187] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer readable media do not include electric carrier signals and telecommunication signals.

[0188] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0189] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for segmenting buildings and roads into individual parts, characterized in that: The steps include: S1. Collect 2-meter multispectral image data of buildings and roads from Gaofen-1, and manually create a dataset of individual building and road labels; S2. Construct deep multi-scale feature extraction module DE_MSCAN; S3. Construct a multi-head attention mechanism module Mhead_BRN for building roads; S4. Based on the deep multi-scale feature extraction module DE_MSCAN constructed in step S2 and the multi-head attention mechanism module Mhead_BRN for building roads constructed in step S3, a feature hierarchical extraction module encoder is constructed; S5. Construct a deep and shallow multi-scale void space pyramid module DS_MASPP, use LightHamHead and MLP methods for decoding, and build a feature decoding module decoder; S6. Design multiple loss functions to guide the optimization of the building-road segmentation model; S7. Based on the feature hierarchical extraction module encoder formed in step S4, the feature decoding module decoder formed in step S5 and the multi-loss function designed in step S6, a building road monomer segmentation model DSMALnet is constructed; S8. Input the building and road monomer label data set produced in step S1 into the building road monomer segmentation model constructed in step S7 for model training to obtain the optimal building road monomer segmentation model; S9. Use the optimal building and road monomer segmentation model obtained in step S8 to segment the buildings and roads in the target area.

2. A method for segmenting buildings and roads according to claim 1, characterized in that: The specific implementation method of step S2 includes the following steps: S2.

1. Setting the DE_MSCAN module consists of two batch normalization modules BN, one multi-head attention mechanism module Mhead_BRN for buildings and roads, and one feed-forward neural network FFN. The calculation process expression is: s DE_MSCAN =Mhead_BRN(BN(input t ))+input t (1) x DE_MSCAN =FFN(BN(s DE_MSCAN )))+s DE_MSCAN (2) Among them, input t Represents input feature information, s DE_MSCAN represents the feature information after BN and Mhead_BRN operations and matrix addition of the input feature information, x DE_MSCAN It means that after BN and FFN operations, it is combined with s DE_MSCAN Feature information for matrix addition, BN(.) represents batch normalization operation, Mhead_BRN(.) represents attention mechanism module calculation, FFN(.) represents FFN module calculation; S2.2.FFN module calculation process expression is: FFN=Conv 1×1 (GELU(Dw_Conv 3×3 (Conv. 1×1 (s DE_MSCAN) )))) (3) Among them, Conv 1×1 (.) indicates a 1×1 convolution operation, GELU(.) indicates an activation function operation, and Dw_Conv 3×3 (.) represents a 3×3 depthwise convolution operation.

3. A method for segmenting buildings and roads according to claim 2, characterized in that: The specific implementation method of step S3 includes the following steps: S3.

1. Set the Mhead-BRN module to consist of two 1×1 convolutions, one GELU activation function, and one multi-attention mechanism Mhead-BR module. The calculation process expression of the Mhead BRN module is: s Mhead _ BRN =GELU(Conv 1×1 (input h ))(4) x Mhead _ BRN =Conv 1×1 (Size—BR(s) Mhead _ BRN ))(5) Among them, s Mhead _ BRN Represents the feature information after 1×1 convolution and GELU activation function operation, x Mhead _ BRN Represents the feature information after the Mhead_BR module and the 1×1 convolution operation, Conv 1×1 (.) indicates 1×1 convolution operation, GELU(.) indicates activation function operation, and Mhead-BR(.) indicates Mhead-BR module calculation; S3.

2. Setting Mhead—BR module consists of 3 parts, the specific structure is as follows: S3.2.

1. The first part is set as a multi-scale spatial and channel attention mechanism, which consists of 7 deep convolutions of different scales. The expression of the calculation process is: s dw5 =Dw-Conv 5×5 (s Mhead_BRN ) (6) Among them, s dw5 、s dw7 、s dw11 、s dw21 Respectively represent the feature information of deep convolution with 5×5, 7×7, 11×11, and 21×21 convolution kernels, Dw_Conv i×j represents the depthwise convolution operation of the i×j convolution kernel, i, j∈[1, 5, 7, 11, 21]; S3.2.

2. Set the second part as a multi-scale position attention mechanism, which consists of 2 average pooling, 2 maximum pooling, 4 normalized attention modules NAM, and 1 1×1 convolution. The expression of the calculation process is: H_NAM,W_NAM=split(Conv 1×1 (concat(s h_d ,s w_d T ))) (10) x pos =H_NAM×W_NAM (11) Among them, s h_avgn ,s h_maxn ,s w_avgn ,s w_maxn They represent the features obtained by performing average pooling in the height direction, maximum pooling in the height direction, average pooling in the width direction, and maximum pooling in the width direction and then performing NAM calculation, respectively. h_d ,s w_d They represent the features obtained by performing matrix dot multiplication on the features in the height and width directions obtained by formula 8, H_NAM and W_NAM represent the features obtained by formula 9 after matrix connection, 1×1 convolution and matrix segmentation along the height and width, xpos represents the weight matrix after matrix multiplication of the features obtained by formula 10, H_AvgPool(.) and W_AvgPool(.) represent the average pooling operations in the height and width directions, H_MaxPool(.) and W_MaxPool(.) represent the maximum pooling operations in the height and width directions, NAM(.) represents the NAM module calculation, Conv 1×1 (.) represents a 1×1 convolution operation, split(.) represents the matrix split operation, T represents the matrix transpose operation, and ⊙ represents the matrix dot multiplication calculation; The NAM calculation formula is as follows: Y(BN)=ω λ (BN(Y(F))) (12) Among them, Y(F) represents the input feature map, λ represents the scale factor of each channel, ω λ Represents the obtained weight, μ B and σ B Represent the mean and standard deviation of the mini-batch, r and β represent the trainable finite transformation scale and offset, BN(s) represents the scale factor, ∈ represents the penalty factor, Bin out Represents the output features of the previous layer; S3.2.

3. Add the four weight matrices obtained in the first part to the one weight matrix obtained in the second part, and then perform a 1×1 convolution operation. The obtained weight matrix is ​​matrix-multiplied with the input feature information. The expression of the calculation process is: x dw =Conv 1×1 (s dw5 +s dw7 +s dw11 +s dw21 +x pos )×s Mhead_BRN (15) Among them, x dw Represents the feature information operated by the Mhead_BRN module.

4. A method for segmenting buildings and roads according to claim 3, characterized in that: The specific implementation method of step S4 is to perform a convolution operation with a 3×3 convolution kernel and a step size of 2 on the input data set to obtain the calculation result Stage0 of the 0th stage; Stage0 is calculated for 4 stages, that is, the DEMSCAN operation designed in step S2 is performed 2 times, 2 times, 4 times, and 2 times. After each stage, a convolution operation with a 3×3 convolution kernel and a step size of 2 is connected for downsampling, and the calculation results of the 4 stages are respectively the calculation result Stage1 of the 1st stage, the calculation result Stage2 of the 2nd stage, the calculation result Stage3 of the 3rd stage, and the calculation result Stage4 of the 4th stage.

5. A method for segmenting buildings and roads according to claim 4, characterized in that: The DSMASPP module of step S5 is divided into two parts, and the specific implementation method is as follows: S5.

1. Setting the first part of the DS MASPP module is shallow multi-scale information fusion, which consists of a 3×3 convolution kernel, a batch normalization BN, and a Relu6 activation function. The expression of the calculation process is: Stage0_1=Relu6(BN(Conv 3×3 (concat(Stage0,bicubic(Stage1))))) (16) Among them, Stage0_1 represents the shallow multi-scale information fusion feature, bicubic(.) represents the bicubic interpolation operation, concat(.,.) represents the matrix connection operation, Conv 3×3 (.) represents a 3×3 convolution operation, BN(.) represents a batch normalization operation, and Relu6(.) represents a Relu6 activation function operation; S5.

2. Setting the second part of the DS MASPP module is the mid-deep multi-scale information fusion, which consists of 3 3×3 dilated convolutions with separation rates of 3, 5, and 7, 4 batch normalization BNs, 4 relu6 activation functions, and 2 3×3 convolutions. The expression of the calculation process is: Stage2_4=Relu6(BN(Conv 3×3 concat(Stage2,bicubic(Stage3,Stage4)))) (17) Stage d =Conv 3×3 (concat(s rate3 ,s rate5 ,s rate7 )) (19) Among them, Stage2_4 represents the features after the middle and deep features are connected, s rate3 ,s rate5 ,s rate7 Respectively represent the feature information after the dilated convolution operation with separation rates of 3, 5, and 7, Stage d Denotes the features after fusion of mid- and deep-level multi-scale information, D_Conv rate=3 、D_Conv rate=5 、D_Conv rate=7 They represent dilated convolution operations with separation rates of 3, 5, and 7, respectively; S5.

3. Combine Stage0_1 obtained in step S5.1 with Stage d Stored in the List list, and then input into LightHamHead and MLP for feature decoding to obtain the predicted building road result y out .

6. A method for segmenting buildings and roads according to claim 5, characterized in that: The specific implementation method of step S6 is to convert the artificial building and road monomer label data ylabe obtained in step S1 into l Multi-classification cross entropy loss is performed on yout obtained in step S5 Calculation, Dice loss function Calculation, edge detection loss function After calculation, the losses are added to obtain a multi-loss function To guide model optimization, the calculation expression is: Multi-classification cross entropy loss The calculation formula is as follows: Where M represents the number of categories; y o,c represents a binary indicator, which is 1 if category c is the correct category of the object and 0 otherwise; po,c represents the probability that the model predicts that the object belongs to category c; Dice loss function The calculation formula is as follows: Among them, F p represents the predicted label set; F gt represents the true label set; Edge Detection Loss Function The calculation formula is as follows: Among them, using the Laplace operator Label y label and the predicted y out For edge extraction, the background is 0 and the edge is 1. The calculation formula is as follows: Among them, f(k,u) is a pixel point in the image, and f(k-1,u), f(k+1,u), f(k,u-1), and f(k,u+1) are the four adjacent pixel points around f(k,u).

7. A method for segmenting buildings and roads according to claim 6, characterized in that: Step S7 constructs the building road monomer segmentation model DSMALnet, which is divided into two stages: encoder and decoder; the encoder is composed of 5 3×3 convolutions for downsampling and 10 DE MSCANs. After the first four downsamplings, 2, 2, 4, and 2 DE MSCAN calculations are performed respectively to realize multi-scale, multi-space, and multi-position feature information extraction; The decoder inputs the feature information of each downsample into the DS_MASPP module, and then inputs it into the lightweight Hamburger and multi-layer perceptron MLP for feature decoding to obtain the prediction result; the loss function uses a multi-loss function to evaluate the model loss.

8. A method for segmenting buildings and roads according to claim 7, characterized in that: In step S8, the building and road monomer label data set produced in step S1 is input into the building road monomer segmentation model DSMALnet constructed in step S7 for model training to obtain the optimal building road monomer segmentation model; the accuracy of the model segmentation is evaluated using the average intersection over union ratio mIoU, category average pixel accuracy mPA, overall classification accuracy Accuracy, accuracy Precision, recall rate Recall, and F1 score.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for segmenting buildings and roads into monomers as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for segmenting buildings and roads into monomers as described in any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Remote sensing image building automatic extraction processing method based on TransUnet

    CN114581770A

  • Instance segmentation method based on detection enhancement and multi-stage bounding box feature refinement

    CN115797629A

  • Comprehensive evaluation method for land classification, electronic equipment and storage medium

    CN117113189A

  • Land classification method based on deep and shallow feature and multi-attention mechanism fusion, electronic equipment and storage medium

    CN117173462A

  • Prostate accidental cancer prediction method and system based on multi-task learning

    CN118172363A

Cited By

  • Identification method for quickly extracting boundary of mango planting area, electronic equipment and storage medium

    CN119832240A

  • A method for quickly extracting a mango planting area boundary, an electronic device, and a storage medium

    CN119832240B