Color steel tile illegal building segmentation detection and geographic information recovery method and system

Abstract

The invention discloses a color steel tile illegal building segmentation detection and geographic information recovery method and system, and relates to the technical field of illegal building detection, and the method comprises the steps: obtaining image data of a training region and a test region, and screening images containing color steel tile illegal buildings to form a color steel tile illegal building data set; marking the color steel tile building, and dividing marked data into training set data and verification set data; carrying out redeployment and channel number adjustment on the global encoder and the local encoder, and constructing an improved neural network model; the color steel tile illegal building data set is input into the improved neural network model for training, and an optimal model for color steel tile illegal building segmentation detection is obtained; the model is used to segment the color steel tile illegal building on the image data of the test area, and the geographic information recovery module is used to recover the geographic information. According to the method, the area and geographic information of the color steel tile illegal building can be obtained, and meanwhile, the advantage of processing a large amount of data is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of illegal building detection, in particular to a method and system for segmenting and detecting color steel tile illegal buildings and restoring geographic information. Background Art

[0002] The renovation of illegal color steel tile buildings in areas such as urban villages and old residential communities is a top priority in promoting urbanization. Such illegal buildings, which are constructed by adding steel frames and color steel tiles to existing houses, not only seriously occupy urban public resources but also pose great safety hazards. Currently, the main methods for identifying such buildings are manual visits, drone aerial photography, and remote sensing image interpretation, etc., with a high degree of dependence on manual labor. How to accurately and automatically obtain the locations of such illegal buildings is crucial for accelerating urban development and the urbanization process.

[0003] In recent years, satellite remote sensing technology has developed rapidly. Currently, the resolution of domestic satellite remote sensing images can reach 1m - 0.5m, but it is still far from being able to detect tiny ground targets. The development of unmanned aerial photography technology has better solved this problem. Its advantages of being unrestricted by geographical location and obtaining centimeter-level images have led to its extensive application in the field of urban illegal building detection. At the same time, object detection based on deep learning has been intelligently implemented in various fields. Its technical advantages lie in its efficient computing and the ability to autonomously learn features and patterns in images. However, traditional deep learning models often have problems such as long training time and high performance requirements for the computing platform. Therefore, there is an urgent need for a deep learning model with a fast training speed, low platform requirements, and excellent performance to achieve rapid detection of color steel tile illegal buildings. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that traditional deep learning models often have problems such as long training time and high performance requirements for the computing platform.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A method for segmenting and detecting color steel tile illegal buildings and restoring geographic information, which includes obtaining image data of the training area and the test area by means of drone aerial photography, screening images containing color steel tile illegal buildings to form a color steel tile illegal building data set; based on the already screened color steel tile illegal building data set, annotating the color steel tile buildings and dividing the annotated data into training set data and validation set data; based on a deep learning model, redeploying and adjusting the number of channels of the global encoder and the local encoder to construct an improved neural network model;

[0007] Based on the improved neural network model and the well-divided training set data, the color steel tile illegal building dataset is input into the improved neural network model for training to obtain the best model for the segmentation and detection of color steel tile illegal buildings; based on the best model for the segmentation and detection of color steel tile illegal buildings, the image data in the test area is segmented for color steel tile illegal buildings, and the geographical information is restored through the geographical information restoration module.

[0008] As a preferred scheme of the method for segmenting and detecting color steel tile illegal buildings and restoring geographical information according to the present invention, wherein: the construction of the improved neural network model includes a five-stage encoder and a four-stage decoder; the five-stage encoder is represented as En1, En2, En3, En4, En5; the four-stage decoder is represented as De1, De2, De3, De4; the RSU modules with different numbers of layers are used in the En1 to En3 stages and the De1 to De3 stages, which are the RSU-7 module, the RSU-6 module, and the RSU-5 module respectively; the RSU-4F module is used in the En4, En5, and De4 stages, and its number of layers is fixed at 4; the number of input data channels I, the number of intermediate channels M, and the number of output data channels O of all RSU modules are optimized and adjusted through the genetic algorithm.

[0009] As a preferred scheme of the method for segmenting and detecting color steel tile illegal buildings and restoring geographical information according to the present invention, wherein: the optimization and adjustment through the genetic algorithm includes initializing the population: the population size is 200 individuals, each individual represents each channel number configuration, the range of the channel number is between [0, 512], and the channel number must be divisible by 2; using the GA slope function as the fitness function, and each channel number configuration is trained on the training dataset for 10 epochs; if a single epoch exceeds 10 minutes, the current channel number scheme is discarded; according to the fitness evaluation result, a selection operation is performed; two individuals are randomly selected from the current population as parents, and new individuals are generated through the crossover operation; the mutation operation is performed on the newly generated individuals, and the mutation rate is set to 0.001; the selection, crossover, and mutation operations are repeated until the maximum number of iterations of 200 times is reached; after each iteration, the population is updated according to the evaluation result of the GA slope function; the individual with the highest fitness in the final generation population is selected as the best channel number configuration.

[0010] As a preferred scheme of the method for segmenting and detecting color steel tile illegal buildings and restoring geographical information according to the present invention, wherein: the GA slope function is expressed as:

[0011]

[0012] wherein, loss 10The loos value of the model when trained to 10 epochs, and loss1 is the loos value of the model when trained to the 1st epoch.

[0013] As a preferred solution of the method for detecting and segmenting illegal color steel tile buildings and restoring geographic information according to the present invention, wherein: inputting the illegal color steel tile building dataset into the improved neural network model for training includes setting training parameters: the number of iterations is set to 100, and the initial learning rate is 0.001; using the Adam optimizer, and the loss function uses binary cross-entropy with logits; completing the warm-up operation at the 2nd epoch, and using the cosine decay strategy to update the learning rate.

[0014] As a preferred solution of the method for detecting and segmenting illegal color steel tile buildings and restoring geographic information according to the present invention, wherein: the specific mathematical expression of the cosine decay strategy is:

[0015]

[0016] where i is the number of training rounds; and respectively represent the maximum and minimum values of the learning rate; T cur represents the number of times of training rounds that have been completed currently; T i represents the total number of epochs in the i-th training.

[0017] As a preferred solution of the method for detecting and segmenting illegal color steel tile buildings and restoring geographic information according to the present invention, wherein: the restoration of geographic information through the geographic information restoration module includes using the best model for detecting and segmenting illegal color steel tile buildings to perform mask extraction on the illegal color steel tile buildings in the test area image patches; inputting the obtained classification results into the geographic information restoration module and splicing them according to the original positions; inputting the test area UAV images into the geographic information restoration module, and using the GDAL package provided by Python to realize the restoration of geographic information for the spliced classification results.

[0018] Another object of the present invention is to provide a system for detecting and segmenting illegal color steel tile buildings and restoring geographic information, which can quickly detect illegal color steel tile buildings.

[0019] To solve the above technical problems, the present invention provides the following technical solutions: A system for a method of detecting and recovering geographic information of illegal color steel tile buildings, comprising: an acquisition module, a labeling module, a segmentation module, an input module, and a geographic information recovery module; the acquisition module acquires image data of a training area and a test area based on an unmanned aerial vehicle (UAV) aerial photography method, and screens images containing illegal color steel tile buildings to form an illegal color steel tile building data set; the labeling module labels the color steel tile buildings based on the screened illegal color steel tile building data set, and divides the labeled data into training set data and validation set data; the segmentation module re-deploys the global encoder and the local encoder and adjusts the number of channels based on a deep learning model to construct an improved neural network model; the input module inputs the illegal color steel tile building data set into the improved neural network model for training based on the improved neural network model and the divided training set data to obtain an optimal model for detecting and segmenting illegal color steel tile buildings; the geographic information recovery module segments the image data of the test area for illegal color steel tile buildings based on the optimal model for detecting and segmenting illegal color steel tile buildings, and performs geographic information recovery through the geographic information recovery module.

[0020] A computer device, comprising a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned method for detecting and recovering geographic information of illegal color steel tile buildings are implemented.

[0021] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method for detecting and recovering geographic information of illegal color steel tile buildings are implemented.

[0022] The beneficial effects of the present invention are as follows: The T-U 2 net proposed by the present invention realizes the automatic detection of illegal color steel tile buildings based on UAV images. Compared with the existing models, the T-U 2 net provided by the present invention can not only process small sample data sets in complex scenarios, but also is superior to the traditional U 2 net, Unet, and Unet++ in terms of training speed and prediction efficiency. The present invention avoids a large number of manual detection processes, and has strong applicability, and can complete detection tasks on UAV images in different regions, which is of great significance for quickly and efficiently obtaining information on illegal color steel tile buildings in cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0024] Figure 1 It is a schematic diagram of the basic process of a method for detecting and segmenting illegal color steel tile buildings and restoring geographic information in Embodiment 1.

[0025] Figure 2 It is a structural diagram of an improved neural network model of a method for detecting and segmenting illegal color steel tile buildings and restoring geographic information in Embodiment 1.

[0026] Figure 3 It is a schematic diagram of the final extracted mask result of illegal color steel tile buildings and the true distribution of a method for detecting and segmenting illegal color steel tile buildings and restoring geographic information in Embodiment 2.

[0027] Figure 4 It is a comparison chart of the present invention and a typical case in terms of GA slope and training time in a method for detecting and segmenting illegal color steel tile buildings and restoring geographic information in Embodiment 2.

[0028] Figure 5 It is a module structure diagram of a system for detecting and segmenting illegal color steel tile buildings and restoring geographic information in Embodiment 3. Specific Embodiments

[0029] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention in conjunction with the drawings in the specification.

[0030] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0031] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a method for detecting and segmenting illegal color steel tile buildings and restoring geographic information, including, as Figure 1 shown:

[0032] S1: Obtain image data of the training area and the test area based on the UAV aerial photography method, and screen the images containing illegal color steel tile buildings to form an illegal color steel tile building dataset.

[0033] In the embodiments of the present application, since the quality of UAV images is easily affected by weather conditions, it is preferable to shoot on a sunny and cloudless day.

[0034] S2: Based on the selected dataset of illegal color steel tile buildings, label the color steel tile buildings, and divide the labeled data into training set data and validation set data.

[0035] In the embodiments of the present application, Agisoft meatshape software is used for the post-processing of UAV aerial photos to generate high-quality training area and test area images, and the output image format is.tif. Based on the Opencv library provided by Python, crop the generated training area and test area images to obtain image blocks of size 512*512; manually screen the numerous small image images generated in the training area and use the Labelme annotation software to label the color steel tile buildings. The labeled data is divided into training and validation datasets according to the ratio of 8:2, and the image blocks generated in the test area are used as test data.

[0036] S3: Based on the deep learning model, redeploy and adjust the number of channels of the global encoder and the local encoder to construct an improved neural network model. As Figure 2 shown in the figure, Skip connection in the figure represents a cross-layer connection, which refers to a structure that bypasses some layers and directly connects the output of one layer to the output of a certain layer behind; down sample represents downsampling, which refers to the process of reducing the spatial or temporal resolution of data; up sample represents upsampling, which refers to the process of increasing the spatial or temporal resolution of data; dilation represents the dilation rate as a hyperparameter; Conv represents a convolutional layer, bn represents a batch normalization layer, and ReLU represents an activation function.

[0037] In the embodiments of the present application, a neural network model is constructed based on U 2 net, and on this basis, its structure is adjusted and the number of channels is finely tuned. The improved neural network model is named T-U 2 net, where the genetic algorithm code and GA slope function are both written in the Python environment. The GA slope function will be used as the fitness function of the genetic algorithm to search for the optimal solution of the number of channels of each module of T-U 2 net. Considering that the increase in network complexity will lead to too long training time, the network configurations with too long training time will be eliminated, that is, if a single epoch exceeds 10 minutes, the channel number scheme will be discarded.

[0038] In the embodiments of the present application, the GA slope function is expressed as:

[0039]

[0040] where loss 10 is the loos value of the model when trained to 10 epochs, and loss1 is the loos value of the model when trained to the 1st epoch.

[0041] Preferably, T-U 2 net optimizes the deployment of the global encoder (RSU (I,M,O) ) and the local encoder (RSU-4F (I,M,O) ) while retaining the original large "U" shape structure. The main network of T-U 2 net consists of a five-stage encoder (En1 to En5) and a four-stage decoder (De1 to De4). In the En1 to En3 and De1 to De3 stages, the corresponding RSU modules are RSU-7 (I,M,O) , RSU-6 (I,M,O) and RSU-5 (I,M,O) respectively; in En4, En5 and De4, the corresponding RSU modules are all RSU-4F (I,M,O) modules.

[0042] Preferably, for RSU-L (I,M,O) (L = 5, 6, 7), L is the number of layers of the RSU module, and the number of layers of the RSU-4F module is 4; for RSU-L (I,M,O) (L = 5, 6, 7, 4F), I is the number of channels of the input data, M is the number of intermediate channels, and O is the number of channels of the output data.

[0043] Preferably, all the number of channels inside T-U 2 net are searched for the optimal number of channels using a genetic algorithm. Each sub-model searched is trained on the training dataset for 10 epochs, and the descent rate of the obtained loss function is used as an index to quantify the quality of each model. At the same time, the input and output channel numbers of each RSU module have a specific ratio relationship to accelerate convergence, that is, the ratio of M to O of the RSU-L and RSU-4F modules is not less than 1 and I is not greater than 3.

[0044] Preferably, the final obtained I of the En1 stage is 3, M is 16, and O is 32; for the En2 to En5 stages, I, M and O are respectively 32, 16 and 32; in the RSU-L (I,M,O) (L = 5, 6, 7, 4F) modules of the De1 to De4 stages, I is 64, M is 16, and O is 32 for all.

[0045] In the embodiments of the present application, T-U 2All the number of channels of the network are randomly initialized in the range of [0, 512], and the initialized number of channels must be divisible by 2. The parameters of the genetic algorithm are set as follows: population size 200, number of iterations 200, and mutation rate 0.001.

[0046] S4: Based on the improved neural network model and the partitioned training set data, the color steel tile illegal building dataset is input into the improved neural network model for training to obtain the best model for the segmentation and detection of color steel tile illegal buildings.

[0047] In the embodiment of this application, the epoch (number of iterations) is set to 100, the initial learning rate is 0.001. By the end of the 2nd epoch, the warmup operation is completed. During the training process, the Adam optimizer is adopted, and the loss function uses binary cross entropy with logits, and the calling method is torch.nn.functional.binary_cross_entropy_with_logits; the initial learning rate of the optimizer is set to 0.001, the weight decay is set to 0.0001, and other parameters remain default; the learning rate update method simultaneously adopts the warmup operation and the Cosine Decay strategy, where the number of cycles of the warmup operation is set to 2.

[0048] In the embodiment of this application, the specific mathematical expression of the Cosine Decay strategy is:

[0049]

[0050] where i is the number of training rounds; and respectively represent the maximum and minimum values of the learning rate, which define the range of the learning rate; T cur represents how many rounds of training have been completed; T i represents the total number of epochs in the i-th training.

[0051] S5: Based on the best model for the segmentation and detection of color steel tile illegal buildings, use this model to segment the image data in the test area for color steel tile illegal buildings, and perform the restoration of geographic information through the geographic information restoration module.

[0052] In the embodiment of this application, the geographic information restoration module (Geo IRM)The functions are implemented based on the GDAL package provided by Python, mainly including the splicing part and the geographical information restoration part. The splicing algorithm splices the numerous mask results of the illegal color steel tile buildings output by the segmentation module according to the original positions, so that the complete test area is covered; the geographical information restoration algorithm restores the geographical information of the spliced mask results of the illegal color steel tile buildings. The operation of the geographical information restoration module is based on the UAV image data, so that the geographical position information of the mask results of the illegal color steel tile buildings in different regions can be restored.

[0053] Preferably, Geo IRM The functions are implemented with the GDAL package provided by Python. The results output by this module are used for subsequent area calculation and special applications.

[0054] Example 2, referring to Figure 3 and Figure 4 This is the second embodiment of the present invention. The difference from the first embodiment is that a method for segmenting and detecting illegal color steel tile buildings and restoring geographical information further includes, in order to verify and explain the technical effects adopted in this method, in this embodiment, a traditional technical solution and the method of the present invention are used for comparative testing, and the test results are compared by means of scientific demonstration to verify the real effects of this method.

[0055] From Figure 3 it can be seen that the finally searched T-U 2 net of the present invention can converge quickly in a short time, which indicates that the reasonable design of the number of channels of the RSU module can balance the model between GAslope and training time, thus greatly improving the efficiency of the deep learning model. Compared with the manually designed U2net, Unet and Unet++, the present invention shows obvious advantages, which will provide an important reference for the further exploration of small deep learning models.

[0056] Example 3, referring to Figure 5, which is the third embodiment of the present invention and is different from the previous two embodiments: A system for the method of detecting the segmentation of color steel tile illegal buildings and restoring geographical information includes an acquisition module 100, a labeling module 200, a segmentation module 300, an input module 400, and a geographical information restoration module 500; the acquisition module 100 acquires the image data of the training area and the test area based on the unmanned aerial vehicle (UAV) aerial photography method, and filters the images containing color steel tile illegal buildings to form a color steel tile illegal building data set; the labeling module 200 labels the color steel tile buildings based on the already filtered color steel tile illegal building data set, and divides the labeled data into training set data and validation set data; the segmentation module 300 redeploys and adjusts the number of channels of the global encoder and the local encoder based on the deep learning model to construct an improved neural network model; the input module 400 inputs the color steel tile illegal building data set into the improved neural network model for training based on the improved neural network model and the divided training set data to obtain the best model for the segmentation detection of color steel tile illegal buildings; the geographical information restoration module 500 segments the color steel tile illegal buildings in the image data of the test area based on the best model for the segmentation detection of color steel tile illegal buildings, and restores the geographical information through the geographical information restoration module.

[0057] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0058] The logic and / or steps represented in the flowchart or otherwise described herein can be considered, for example, a definable sequence list of executable instructions for implementing a logical function, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or in conjunction with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0059] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0060] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0061] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting the segmentation of illegal color steel tile buildings and restoring geographical information, characterized in that: including, acquiring image data of the training area and the test area by means of UAV aerial photography, and screening images containing color steel tile illegal buildings to form a color steel tile illegal building data set; based on the screened color steel tile illegal building data set, annotating the color steel tile buildings, and dividing the completed annotated data into training set data and validation set data; based on the deep learning model, redeploying the global encoder and the local encoder and adjusting the number of channels, and constructing an improved neural network model; based on the improved neural network model and the divided training set data, inputting the color steel tile illegal building data set into the improved neural network model for training to obtain the best model for color steel tile illegal building segmentation detection; based on the best model for color steel tile illegal building segmentation detection, segmenting the color steel tile illegal buildings in the image data of the test area, and restoring the geographical information through the geographical information restoration module.

2. The method for detecting the segmentation of illegal color steel tile buildings and restoring geographic information according to claim 1, characterized in that: The construction of the improved neural network model includes a five-stage encoder and a four-stage decoder; The five-stage encoder is represented as En1, En2, En3, En4, En5; The four-stage decoder is shown as De1, De2, De3, De4; The RSU modules with different numbers of layers are used in the En1 to En3 stages and the De1 to De3 stages, which are the RSU-7 module, the RSU-6 module, and the RSU-5 module respectively; The RSU-4F module is used in the En4, En5, and De4 stages, and its number of layers is fixed at 4; The number of input data channels I, the number of intermediate channels M, and the number of output data channels O of all RSU modules are optimized and adjusted through the genetic algorithm.

3. A method for detecting the segmentation of illegal color steel tile buildings and restoring geographical information according to claim 2, characterized in that: The optimization and adjustment through the genetic algorithm includes, initializing the population: the population size is 200 individuals, each individual represents each channel number configuration, the range of the channel number is between [0, 512], and the channel number must be divisible by 2; Using GA slope as the fitness function, each channel number configuration is trained for 10 epochs on the training dataset; if a single epoch exceeds 10 minutes, discard the current channel number scheme; performing a selection operation according to the fitness evaluation result; randomly selecting two individuals from the current population as parents, and generating new individuals through crossover operation; performing a mutation operation on the newly generated individuals, and setting the mutation rate to 0.001; repeating the selection, crossover, and mutation operations until the maximum number of iterations of 200 times is reached; After each iteration, update the population according to the evaluation result of the GA slope function; selecting the individual with the highest fitness in the final generation population as the best channel number configuration.

4. The method for detecting the segmentation of color steel tile illegal buildings and restoring geographical information according to claim 3, wherein: The GA slope function is expressed as: Among them, loss 10 is the loos value of the model when trained to 10 epochs, and loss1 is the loos value of the model when trained to the 1st epoch.

5. The method for detecting the division of color steel tile illegal buildings and restoring geographical information according to claim 4, characterized in that: The inputting the color steel tile illegal building data set into the improved neural network model for training includes, setting training parameters: the number of iterations is set to 100, and the initial learning rate is 0.001; adopting the Adam optimizer, and using the binary cross-entropy with logits as the loss function; completing the warm-up operation at the 2nd epoch, and adopting the cosine decay strategy to update the learning rate.

6. The method for detecting the segmentation of illegal color steel tile buildings and restoring geographical information according to claim 5, characterized in that: The specific mathematical expression of the cosine decay strategy is: Among them, i is the number of training rounds; and respectively represent the maximum and minimum values of the learning rate; T cur represents the number of times of training for the current completed rounds; T i represents the total number of epochs in the i-th training.

7. A method for detecting the segmentation of illegal color steel tile buildings and restoring geographical information according to claim 6, characterized in that: The restoration of the geographical information through the geographical information restoration module includes, using the best model for color steel tile illegal building segmentation detection to perform mask extraction on the color steel tile illegal buildings in the test area image patches; transmitting the obtained classification results to the geographical information restoration module and splicing them according to the original positions. The UAV images of the test area are transmitted to the geographic information restoration module, and the GDAL package provided by Python is used to restore the geographic information of the spliced classification results.

8. A system adopting a color steel tile illegal building segmentation detection and geographic information restoration method as described in any one of claims 1 to 7, characterized in that: It includes an acquisition module (100), an annotation module (200), a segmentation module (300), an input module (400), and a geographic information restoration module (500); The acquisition module (100) obtains the image data of the training area and the test area based on the UAV aerial photography method, and screens the images containing color steel tile illegal buildings to form a color steel tile illegal building data set; The annotation module (200) annotates the color steel tile buildings based on the screened color steel tile illegal building data set, and divides the annotated data into training set data and validation set data; The segmentation module (300) re-deploys the global encoder and the local encoder and adjusts the number of channels based on the deep learning model to construct an improved neural network model; The input module (400) inputs the color steel tile illegal building data set into the improved neural network model for training based on the improved neural network model and the divided training set data to obtain the best model for the segmentation and detection of color steel tile illegal buildings; The geographic information restoration module (500) segments the color steel tile illegal buildings in the image data of the test area based on the best model for the segmentation and detection of color steel tile illegal buildings, and restores the geographic information through the geographic information restoration module.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for segmenting and detecting color steel tile illegal buildings and restoring geographic information according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for segmenting and detecting color steel tile illegal buildings and restoring geographic information according to any one of claims 1 to 7.

Images