System, method, electronic equipment and storage medium for extracting morphological features of erosion trenches

Abstract

This invention relates to the field of remote sensing image processing technology, and particularly to a system, method, electronic device, and storage medium for extracting erosion gully morphological features. The encoder, based on the Xception model backbone network, encodes features from the input erosion gully image to obtain shallow and deep features. The aggregator extracts multiple types of features from the deep features using deformable convolution and depth-separable dilated convolution, and then aggregates the extracted features to obtain fused features. The decoder fuses the fused features with the shallow features, and then decodes the fused features to obtain the erosion gully identification result. This invention is based on the Deeplabv3+ model, fusing erosion gully morphological features as a process mechanism knowledge-driven improvement model to achieve high-precision, automated erosion gully extraction.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing image processing technology, and particularly relates to a system, method, electronic device and storage medium for extracting the morphological features of erosion trenches. Background Technology

[0002] Extraction and spatial mapping of erosion gullies are crucial steps in their integrated management. Currently, remote sensing technology is widely used in erosion gully identification. Traditional visual interpretation methods, which involve manually marking erosion gullies in satellite imagery, are time-consuming and labor-intensive. Machine learning methods require multi-source data and manually designed feature variables, demanding specialized knowledge from the user.

[0003] In recent years, semantic segmentation models based on deep learning have become the mainstream technology for automated extraction of erosion gullies. Deep learning network models can automatically extract the features of the learning target from manually labeled datasets, achieving automated extraction. Convolutional Neural Network (CNN) models, such as U-Net, U-Net++, and the Deeplab series, are most widely used in the field of automated erosion gully extraction. However, due to the irregular, tortuous, and varied complex geometric shapes of erosion gullies, coupled with interference from surrounding features with similar characteristics (streams, rivers, agricultural roads, etc.), it is difficult to achieve accurate automated extraction of erosion gullies.

[0004] In summary, current remote sensing-based methods have three main shortcomings:

[0005] (1) Visual interpretation is inefficient and cannot achieve large-scale extraction of erosion trenches;

[0006] (2) Machine learning methods have high requirements for expert knowledge, multi-source remote sensing data types and resolutions, as well as environmental auxiliary data, and cannot achieve automated extraction;

[0007] (3) Current deep learning network models do not take into account the morphological features of erosion trenches, resulting in low recognition efficiency. Summary of the Invention

[0008] In view of this, the present invention aims to provide a method, system, electronic device and storage medium for extracting the morphological features of erosion trenches. Based on the Deeplabv3+ model, and considering the slender structure, tortuous shape and diverse features of erosion trenches, deformable convolution and depth-separable dilated convolution are introduced into the aggregator, which is more suitable for erosion trench extraction tasks. Since the erosion trenches have complex shapes and many features to learn, Xception, which has a larger number of parameters and better learning effect, is used as the backbone network of the model, in conjunction with the aggregator, to achieve accurate and automated extraction of erosion trenches.

[0009] A system for extracting morphological features of erosion trenches includes: an encoder, a backbone network based on the Xception model, which encodes features of the input erosion trench image to obtain shallow and deep features; an aggregator that performs multiple types of convolutions on the deep features simultaneously to obtain multi-type features; global context information in the deep features; fusion convolution of the obtained global context information and the shallow features to obtain fused features; and a decoder that fuses the fused features and the shallow features, and then decodes the fused features to obtain the erosion trench identification result.

[0010] Furthermore, the backbone network includes: an inlet stream, which performs standard convolution and multiple depth-separable convolutions on the erosion trench image to obtain shallow features; an intermediate stream, which performs multiple depth-separable dilated convolution operations and progressive downsampling on the shallow features to obtain intermediate features; and an outlet stream, which performs depth-separable convolution operations on the intermediate features to obtain deep features.

[0011] Furthermore, 1×1 convolutions are performed on the deep features to obtain global contextual information.

[0012] Furthermore, in the decoder: the fused features are aligned with the shallow features in terms of dimensions and then concatenated in dimensions. The concatenated features are then subjected to depthwise separable convolution to complete feature fusion. The fused features are then subjected to classification convolution and upsampling to obtain the erosion trench identification result.

[0013] Furthermore, a 1×1 convolution is performed on the fused features to complete the classification convolution.

[0014] A method for extracting the morphological features of erosion gullies, comprising:

[0015] S1: Obtain the gully image dataset, divide the land features in each image of the gully image dataset into two categories: gullies and background, generate gully mask images, and obtain a training set containing gully images and corresponding mask images;

[0016] S2: Construct an extraction system for the morphological features of erosion grooves as provided in this invention, and train the extraction system using the training set obtained in step S1 to obtain an extraction model; during the training process, the erosion groove image is used as the input of the extraction model, and the corresponding mask image is used as the output of the extraction model.

[0017] S3: Input the image of the erosion gully to be identified into the extraction model obtained in step S2, and the extraction model outputs the erosion gully identification result.

[0018] Furthermore, in step S2, the loss function obtained by weighted summation of Dice loss and CE loss is used as the loss function during the training process.

[0019] A computer device, comprising:

[0020] At least one processor; and

[0021] A memory that is communicatively connected to at least one processor; wherein,

[0022] The memory stores instructions that can be executed by at least one processor, which enables the at least one processor to perform the method for extracting erosion groove morphology features provided by the present invention.

[0023] A non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method for extracting erosion groove morphology features provided by the present invention.

[0024] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0025] This invention creates a system, method, electronic device, and storage medium for extracting the morphological features of erosion gullies. Combining the distribution and morphological characteristics of black soil erosion gullies, a deformable convolution is introduced into the aggregator to couple the shape of the erosion gullies, focusing on extracting the erosion gully regions in the image. Simultaneously, depthwise separable dilated convolution is used to reduce the number of computational parameters and improve computational efficiency. The aggregator, combined with the Xception backbone network, forms a multi-scale feature fusion network specifically designed for black soil erosion gully extraction. This network can take into account both local detailed features (such as edges, width, and morphological variations) and global structural information (such as distribution, direction, and adjacency relationships) in erosion gully segmentation tasks, significantly improving the segmentation accuracy and efficiency of black soil erosion gullies. It ensures the accuracy and stability of automated erosion gully extraction in complex environments, breaking through the limitations of high-resolution remote sensing images such as those from high-resolution satellites. Attached Figure Description

[0026] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0027] Figure 1 A schematic diagram of the framework of the erosion trench morphology feature extraction system described in the embodiments of the present invention;

[0028] Figure 2 A schematic diagram of the aggregator described in an embodiment of the present invention;

[0029] Figure 3 A flowchart illustrating the method for extracting the morphological features of erosion trenches as described in the embodiments of the present invention;

[0030] Figure 4 A schematic flowchart of the method for extracting the morphological features of erosion trenches as described in the embodiments of the present invention;

[0031] Figure 5 A schematic diagram illustrating the image enhancement process described in an embodiment of the present invention;

[0032] Figure 6 A schematic diagram of the structure of the electronic device described in the embodiment of the present invention.

[0033] Explanation of reference numerals in the attached figures:

[0034] 1. Electronic device; 2. External device; 3. Processing unit; 4. Bus; 5. Network adapter; 6. Display; 7. (I / O) interface; 8. System memory; 9. Random access memory; 10. Cache memory; 11. Storage system; 12. Utility; 13. Program module. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0036] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0037] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0038] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0039] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0040] like Figure 1 As shown in the embodiment of this invention, the erosion groove morphology feature extraction system adopts a Deeplabv3+ structure, specifically including an encoder, an aggregator, and a decoder. The Xception backbone network has a large number of parameters and strong feature extraction capabilities, enabling the extraction of both shallow and deep features. Therefore, the encoder in this invention uses the Xception model; that is, the encoder, based on the Xception model backbone network, performs feature encoding on the input erosion groove image to obtain shallow and deep features. Shallow features mainly carry local information of the image, including the edges, texture, and shape of the erosion groove, while deep features carry global information of the target, obtaining the overall structure and layout of the erosion groove.

[0041] In some embodiments, the backbone network includes an inlet stream, an intermediate stream, and an outlet stream. The inlet stream performs standard convolution and multiple depthwise separable convolutions on the erosion trench image to obtain shallow features. The intermediate stream performs multiple depthwise separable dilated convolution operations and progressive downsampling on the shallow features to obtain intermediate features. The outlet stream performs depthwise separable convolution operations on the intermediate features to obtain deep features.

[0042] In this embodiment of the invention, the ingress stream comprises two 3×3 standard convolutional layers and two depthwise separable convolutions, forming Block 1 and Block 2. The output feature of Block 2 is a shallow feature. The intermediate stream consists of 16 identical convolutional blocks (Block 3-Block 18), each using a depthwise separable dilated convolution. The output feature of Block 18 is an intermediate feature. The egress stream is Block 19, which uses depthwise separable convolutions. The output feature of Block 19 is a depth feature.

[0043] The logic of depthwise separable dilated convolution is as follows: it introduces a dilation parameter into depthwise convolution, transforming it into a depthwise dilated convolution. Depthwise separable dilated convolution first expands the receptive field in the spatial dimension (decoupling spatial correlation) through depthwise dilation, and then adjusts the feature dimension in the channel dimension through pointwise convolution (decoupling channel correlation), thus achieving lightweight design while maintaining a large receptive field. By setting different dilation rates, depthwise separable dilated convolution can expand the receptive field of a depthwise dilated convolution at a certain scale to that of a regular convolution at other scales (e.g., with a dilation rate of 2, the receptive field of a 3×3 depthwise dilated convolution is equivalent to a 5×5 regular convolution; with a dilation rate of 4, it is equivalent to a 9×9 regular convolution), while the number of parameters remains at the level of the convolution scale, and downsampling is not required (i.e., feature map size is preserved), resulting in no resolution loss. Furthermore, depthwise separable dilated convolution inherits the low computational cost of depthwise separable convolution, further improving the lightweight nature of the network model.

[0044] The backbone network progressively downsamples the input image through convolutions, reducing its dimensionality to H / 2×W / 2, H / 4×W / 4, and H / 8×W / 8, where H and W represent the height and width of the image or feature, respectively. Shallow features are extracted at the H / 4×W / 4 stage to preserve details, while deeper features are downsampled to H / 8×W / 8 to obtain the maximum receptive field and extract high-level semantic information from erosion grooves. The backbone network compresses the dimensionality of the feature maps, reducing computational overhead while enhancing the ability to extract detailed features.

[0045] The aggregator extracts multiple feature types from deep features using deformable convolution and depthwise separable dilated convolution, and then aggregates these features to obtain fused features. The aggregator processes deep features while preserving their original semantic recognition capabilities, resulting in output fused features containing richer contextual information.

[0046] In some embodiments, aggregators such as Figure 2 As shown, multiple types of convolutions are performed simultaneously on deep features to obtain multi-type features; global context information is obtained from the deep features; the obtained global context information and the global context information are then fused by convolution to obtain fused features. Specifically, a 1×1 convolution is performed on the deep features to obtain global context information.

[0047] In this embodiment of the invention, the aggregator includes five parallel convolutional branches:

[0048] Branch 1 uses a 1×1 standard convolution to preserve the original receptive field and extract local features;

[0049] Branch 2 uses depthwise separable dilated convolution. The depthwise separable design greatly reduces the number of parameters. With a dilation rate of 12, it provides a receptive field of approximately 25×25, filling the gap between small and large receptive fields.

[0050] Branch 3 uses a standard dilated convolution with a dilation rate of 24 to obtain a wide range of contextual information;

[0051] Branch 4 is a branch that uses variable convolution based on the shape characteristics of the erosion gully. Deformable convolution can dynamically adjust the shape and position of the convolution kernel according to the learned offset, better fit the boundary of the erosion gully, and perform denser sampling in important areas (such as the boundary).

[0052] Branch 5 is a global context branch consisting of 1×1 convolutions. This branch obtains statistical information of the entire feature map to help determine the overall scene type.

[0053] After deep features are processed through 5 parallel branches, 5 feature maps of different scales are generated. The feature maps are then concatenated according to channels and fused using 1×1 convolution in the module to reduce the number of channels and learn the importance weights of each branch. Then, 3×3 convolution is used to further integrate information in space and enhance local consistency.

[0054] The decoder fuses the fused features and shallow features, and then decodes the fused features to obtain the erosion trench identification results.

[0055] In some embodiments, the feature processing in the decoder includes: aligning the fused features with the shallow features in terms of dimensions and then concatenating them in dimensions; then performing a depthwise separable convolution on the concatenated features to complete feature fusion; and finally performing a classification convolution on the fused features and then upsampling to obtain the erosion trench recognition result. Specifically, the classification convolution is implemented using a 1×1 convolution, that is, performing a 1×1 convolution on the fused features to complete the classification convolution.

[0056] In this embodiment of the invention, the feature processing in the decoder includes: performing upsampling of the fused features using bilinear interpolation at twice the original size, resulting in a feature dimension consistent with the shallow features after 1×1 convolution dimensionality reduction, thus aligning the fused features with the shallow features; concatenating the two dimension-aligned features, resulting in a higher-dimensional feature, first using a 3×3 depthwise separable convolution with Dropout 0.5 for strong feature regularization, completing feature dimensionality reduction while preventing overfitting; then using another 3×3 depthwise separable convolution with Dropout 0.1 for weak feature regularization, further integrating features and enhancing feature expressive power while further preventing overfitting. The fused features are then subjected to a 1×1 classification convolution to reduce the feature map channels to 2 according to the erosion groove and background, and finally, a 4x bilinear interpolation upsampling is performed to restore the image size, resulting in an erosion groove recognition result image with the same size as the input erosion groove image.

[0057] In the system provided by this invention, the encoder and aggregator module are smoothly connected. The upsampling operation of deep features in the decoder ensures good cooperation between shallow and deep features while maintaining the efficiency of the model, reducing the omission of small-scale erosion grooves, and ensuring the consistency and completeness of the erosion groove extraction results.

[0058] This invention also provides a method for extracting the morphological features of erosion gullies, combined with... Figure 3 and Figure 4 ,include:

[0059] S1: Obtain the erosion gully image dataset. Divide the land features in each image of the erosion gully image dataset into two categories: erosion gullies and background. Generate a mask image of the erosion gullies to obtain a training set containing erosion gully images and corresponding mask images.

[0060] In this embodiment of the invention, a dataset containing 1034 erosion gully images was constructed. All images were normalized to 256×256 pixels, and the land features in the images were divided into two categories—erosion gullies and background—using a binary method to create a mask image of the erosion gullies. To enhance the diversity of the dataset, this embodiment of the invention also applied methods such as... Figure 5 The spatial augmentation methods shown (rotation, flipping, scaling) are used to randomly generate two images. After this process, the dataset is expanded to 3102 valid images, which are then divided into training, validation, and test sets in an 8:1:1 ratio.

[0061] S2: Construct an extraction system for erosion groove morphology features as provided in the embodiments of the present invention, and train the extraction system using the training set obtained in step S1 to obtain an extraction model; during the training process, the erosion groove image is used as the input of the extraction model, and the corresponding mask image is used as the output of the extraction model.

[0062] In some embodiments, step S2 uses a weighted sum of Dice loss and CE loss as the loss function during training. In typical imbalanced land cover segmentation tasks such as gully extraction, gullies often occupy a very small area in the image, have fragmented shapes, and blurred boundaries. CE loss ensures the model learns global semantic discrimination ability, but using CE loss alone can easily lead to the model ignoring gully regions. Dice loss, on the other hand, drives the model to focus on improving the spatial integrity and boundary fit of the gully region, but using Dice loss alone may lead to training instability. Therefore, this embodiment of the invention combines the two in a weighted manner, effectively integrating their advantages. This combined strategy significantly enhances the robustness of extracting weak edges, small gullies, and discontinuous morphological structures, ultimately achieving high-precision boundary delineation while better maintaining the connectivity of gully morphological structures and the accuracy of geographical attributes.

[0063] In this embodiment of the invention, training begins with pre-weighted Xception backbone network provided by the official documentation, without freezing the backbone network. During training, the loss function is calculated using the validation set, and a cosine annealing (COS) decaying dynamic learning rate scheduling strategy and an SGD gradient descent optimizer are used to address challenges in erosion trench extraction, such as extreme class imbalance, difficulty in learning small targets, and high boundary requirements, guiding the model to converge more stably and accurately.

[0064] S3: Input the image of the erosion gully to be identified into the extraction model obtained in step S2, and the extraction model outputs the erosion gully identification result.

[0065] The method provided in this invention improves upon the Deeplabv3+ model, demonstrating higher segmentation accuracy and robustness in the task of extracting erosion gullies in black soil, especially in complex background environments, effectively maintaining the consistency of segmentation results. This invention achieves multi-scale feature extraction and fusion through an aggregator, accurately capturing the details and global structure of erosion gullies, ensuring high-precision identification in complex farmland scenarios.

[0066] Accordingly, according to embodiments of the present invention, the present invention also provides a computer device, a readable storage medium, and a computer program product.

[0067] Figure 6 This is a schematic diagram of the structure of an electronic device 1 provided in an embodiment of the present invention. Figure 6 A block diagram is shown of an exemplary electronic device 1 suitable for implementing embodiments of the present invention. Figure 6 The electronic device 1 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0068] like Figure 6 As shown, electronic device 1 is represented in the form of a general-purpose computing device. Electronic device 1 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 1 can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0069] The components of electronic device 1 may include, but are not limited to: one or more processors or processing units 3, system memory 8, and bus 4 connecting different system components (including system memory 8 and processing unit 3).

[0070] Bus 4 represents one or more of several bus architectures, including memory buses or memory controllers, peripheral buses, graphics acceleration ports, processors, or local buses using any of the various bus architectures. Examples of these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0071] Electronic device 1 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 1, including volatile and non-volatile media, removable and non-removable media.

[0072] System memory 8 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 9 and / or cache memory 10. Electronic device 1 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 11 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 6 Not shown; usually referred to as a "hard drive"). Although Figure 6 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 4 via one or more data media interfaces. System memory 8 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0073] A program / utility 12 having a set (at least one) of program modules 13 may be stored, for example, in system memory 8. Such program modules 13 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 13 typically perform the functions and / or methods described in the embodiments of the present invention.

[0074] Electronic device 1 can also communicate with one or more external devices 2 (e.g., keyboard, pointing device, display 6, etc.), and with one or more devices that enable a user to interact with electronic device 1, and / or with any device that enables electronic device 1 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 7. Furthermore, electronic device 1 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 5. Figure 6 As shown, network adapter 5 communicates with other modules of electronic device 1 via bus 4. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with electronic device 1, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0075] The processing unit 3 executes various functional applications and data processing by running programs stored in the system memory 8, such as implementing the method for extracting erosion groove morphology features provided in the embodiments of the present invention.

[0076] This invention also provides a non-transitory computer-readable storage medium storing computer instructions, on which a computer program is stored, wherein the program, when executed by a processor, is used to extract the morphological features of erosion trenches provided in all embodiments of this application.

[0077] The computer storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0078] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0079] The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. The computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0080] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the extraction method based on the above-described erosion trench morphological features.

[0081] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0082] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A system for extracting the morphological features of erosion trenches, characterized in that, include: The encoder, based on the backbone network of the Xception model, performs feature encoding on the input erosion trench image to obtain shallow and deep features; The aggregator consists of five parallel convolutional branches: branch one uses 1×1 standard convolution, branch two uses depth-separable dilated convolution, branch three uses standard dilated convolution, branch four uses variable convolution based on the shape of the erosion trench, and branch five performs 1×1 convolution on deep features to obtain global context information. After deep features are processed through 5 parallel branches, 5 feature maps of different scales are generated. The feature maps are then concatenated according to channels and fused using convolution to obtain fused features. The decoder fuses the fused features and shallow features, and then decodes the fused features to obtain the erosion trench identification results. The feature processing in the decoder includes: upsampling the fused features by double bilinear interpolation, so that the dimension of the processed features is consistent with the dimension of the shallow features after 1×1 convolution dimensionality reduction; concatenating the two dimensionally aligned features, performing strong regularization on the concatenated features using a 3×3 depthwise separable convolution, and then performing weak regularization on the features using another 3×3 depthwise separable convolution to further integrate the features; using a 1×1 classification convolution to reduce the feature map channels to 2 according to the erosion groove and the background, and finally performing 4x bilinear interpolation upsampling to restore the image size, resulting in an erosion groove recognition result image with the same size as the input erosion groove image.

2. The system for extracting the morphological features of erosion trenches according to claim 1, characterized in that, The backbone network includes: The inlet flow is used to perform standard convolution and multiple depthwise separable convolutions on the erosion trench image to obtain shallow features; The intermediate flow performs multiple depth-separable dilated convolution operations and progressive downsampling on shallow features to obtain intermediate features. The outflow is used to perform depthwise separable convolution operations on the intermediate features to obtain deep features.

3. The system for extracting the morphological features of erosion trenches according to claim 1, characterized in that, Perform a 1×1 convolution on the fused features to complete the classification convolution.

4. A method for extracting the morphological features of erosion gullies, characterized in that, include: S1: Obtain the gully image dataset, divide the land features in each image of the gully image dataset into two categories: gullies and background, generate gully mask images, and obtain a training set containing gully images and corresponding mask images; S2: Construct an extraction system for the morphological features of erosion trenches as described in any one of claims 1 to 3, and train the extraction system using the training set obtained in step S1 to obtain an extraction model; during the training process, the erosion trench image is used as the input of the extraction model, and the corresponding mask image is used as the output of the extraction model. S3: Input the image of the erosion gully to be identified into the extraction model obtained in step S2, and the extraction model outputs the erosion gully identification result.

5. The method for extracting the morphological features of erosion trenches according to claim 4, characterized in that, In step S2, the loss function obtained by weighted summation of Dice loss and CE loss is used as the loss function during the training process.

6. A computer device, characterized in that, include: At least one processor; as well as A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions executable by at least one processor, which enable the at least one processor to perform the method for extracting the morphological features of the erosion trenches as described in any one of claims 4 to 5.

7. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, Computer instructions are used to cause a computer to execute the method for extracting the morphological features of erosion trenches as described in any one of claims 4 to 5.

Images