A surveying and mapping image enhancement system based on unmanned aerial vehicle remote sensing technology and its implementation method

Abstract

The present invention discloses a surveying and mapping image enhancement system based on unmanned aerial vehicle remote sensing technology and its implementation method. The steps include: S1, autonomously flying and collecting data by carrying optical and thermal imaging sensors on unmanned aerial vehicles; S2, denoising, geometric correction and radiation correction of pre-processed data; S3, using image enhancement algorithms to perform contrast stretching, histogram equalization and edge detection; S4, extracting terrain features from the enhanced image, generating a three-dimensional digital elevation model and a digital terrain map; S5, performing multi-source data fusion, integrating high-precision terrain data into a geographic information system, and supporting map making and spatial analysis. The present invention improves the image resolution and surveying and mapping accuracy of complex terrain areas through multi-sensor fusion and image processing technology, provides detailed and accurate terrain information, supports a variety of applications such as environmental monitoring, urban planning and disaster assessment, and has practical value.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing technology and geographic information system, and in particular to a surveying and mapping image enhancement system based on unmanned aerial vehicle remote sensing technology and an implementation method thereof. Background Art

[0002] In the field of modern surveying and mapping and geographic information systems, UAV remote sensing technology has become an important tool and means. With the development of UAV technology, UAV remote sensing technology is increasingly used in many fields such as environmental monitoring, agriculture, urban planning and disaster management. UAV remote sensing technology can quickly and inexpensively obtain high-resolution geographic information data, promote resource integration, improve surveying and mapping accuracy, and promote technological innovation. However, UAV remote sensing technology also faces many challenges in its application, especially in data processing and fusion. Traditional UAV remote sensing data processing mainly includes data acquisition, preprocessing, image enhancement, feature extraction and three-dimensional terrain modeling. Although existing technical methods have improved the quality and accuracy of remote sensing data to a certain extent, there are still many defects and challenges, which are specifically manifested in the following aspects:

[0003] The first is that multi-sensor data fusion is not perfect. Existing technologies usually rely on a single sensor to obtain data, which limits the integrity and accuracy of the data. Optical sensors, thermal imaging sensors, and LiDAR sensors each have their own advantages, but how to effectively fuse the data of these sensors, make full use of the advantages of different sensors, and improve the mapping accuracy of complex terrain areas is still a technical challenge.

[0004] The second is that the image processing algorithm needs to be optimized. Traditional image processing algorithms often do not work well when processing remote sensing data in complex environments. For example, when the lighting conditions vary greatly, algorithms such as histogram equalization and contrast stretching are difficult to stably improve image quality. In addition, edge detection algorithms are prone to false detection or missed detection when processing images with high noise.

[0005] The third is the lack of accuracy in 3D terrain modeling. Existing 3D terrain modeling technology often lacks accuracy when dealing with complex terrain. Especially when generating digital elevation models and digital terrain maps, how to combine enhanced images and LiDAR data to achieve high-precision 3D modeling is an urgent problem to be solved.

[0006] Fourth, GIS data integration and application are limited. How to efficiently integrate high-precision terrain data into GIS and give full play to its value in various application scenarios is still a challenge. Existing technologies have problems such as low efficiency and insufficient accuracy in data integration and application, which makes it difficult to meet the actual needs of environmental monitoring, urban planning and disaster assessment.

[0007] Therefore, providing a surveying and mapping image enhancement system based on UAV remote sensing technology and an implementation method thereof is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose a surveying and mapping image enhancement system based on unmanned aerial vehicle remote sensing technology and its implementation method. The present invention uses unmanned aerial vehicles equipped with optical sensors and thermal imaging sensors to conduct autonomous flight and remote sensing data collection in the target area, and uses advanced image processing and fusion algorithms to process the collected data. The specific method includes the steps of denoising, geometric correction, radiation correction, contrast stretching, histogram equalization, edge detection, and three-dimensional terrain modeling. Through the integration of multi-sensor data fusion and high-precision terrain data into the geographic information system, the present invention significantly improves the image resolution and surveying and mapping accuracy in complex terrain areas, and has the advantages of providing detailed and accurate terrain information and supporting a variety of applications such as environmental monitoring, urban planning, and disaster assessment.

[0009] According to an embodiment of the present invention, a mapping image enhancement system based on UAV remote sensing technology and an implementation method thereof include the following steps:

[0010] S1. Use the UAV platform to carry optical sensors, LiDAR and thermal imaging sensors to conduct autonomous flight and remote sensing data collection in the target area;

[0011] S2. Preprocessing the collected remote sensing data, including denoising, geometric correction and radiation correction;

[0012] S3, using image enhancement algorithm to perform contrast stretching, histogram equalization and edge detection on the preprocessed image;

[0013] S4, extract important terrain features from the enhanced image and generate high-precision three-dimensional digital elevation model and digital terrain map in combination with LiDAR data;

[0014] S5. Perform multi-source data fusion on data from different sensors and integrate the processed high-precision terrain data into the geographic information system to support map making and spatial analysis.

[0015] Optionally, the S1 specifically includes:

[0016] S11. Select and configure the UAV platform to enable it to carry optical sensors, LiDAR sensors, and thermal imaging sensors;

[0017] S12, setting the flight path and parameters, including flight altitude, speed and route overlap;

[0018] S13, start the UAV, perform autonomous flight along a predetermined path, and synchronously collect optical images, LiDAR data, and thermal imaging data during the flight;

[0019] S14, the collected optical image data is recorded as I optical , LiDAR data is denoted as L LiDAR , thermal imaging data is recorded as T thermal ;

[0020] S15, real-time monitoring of the UAV flight status and sensor working conditions;

[0021] S16. After the flight is completed, recycle the drone and optical , L LiDAR and T thermal Transmitted to ground processing station.

[0022] Optionally, the S2 specifically includes:

[0023] S21, using median filtering or non-local mean filtering to collect the optical image data I optical Perform denoising processing;

[0024] S22, to I optical Perform geometric correction to correct the image geometric distortion caused by the change of the UAV's flight attitude;

[0025] S23, to I optical Perform radiation correction to correct radiation errors caused by changes in lighting conditions;

[0026] S24, collect the LiDAR data L LiDAR Perform denoising to remove noise from LiDAR data;

[0027] S25, for L LiDAR Perform geometric correction;

[0028] S26. Use appropriate filters to eliminate thermal imaging data T thermal Noise in

[0029] S27, T thermal Perform radiation correction to correct radiation errors in thermal imaging data.

[0030] Optionally, the S3 specifically includes:

[0031] S31, preprocessing the optical image data I optical Perform contrast-limited adaptive histogram equalization;

[0032] S32, to I opticalAn improved multi-scale Retinex process is performed to balance the grayscale distribution of the image. The improved multi-scale Retinex process formula is as follows:

[0033] P final (x,y)=∑ s w(s)·(log(I optical (x,y))-log(F(x,y,s)));

[0034] Among them, R final (x,y) represents the fused Retinex output at position ((x,y), w(s) represents the weight of scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s;

[0035] S33, using the improved Canny edge detection algorithm, I optical Perform edge detection to extract edge information from the image. The improved Canny edge detection algorithm formula is as follows:

[0036]

[0037] Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the image in the x and y directions respectively;

[0038] S34, the pre-processed LiDAR data L LiDAR Performing improved adaptive filtering processing;

[0039] S35, using improved multi-scale Retinex processing and adaptive edge detection algorithm, the pre-processed thermal imaging data T thermal Perform contrast enhancement and edge detection.

[0040] Optionally, the S32 specifically includes:

[0041] S321, preprocessing the optical image data I optical Use Gaussian filters of different scales for blurring to obtain a multi-scale blurred image F(x, y, s);

[0042] S322. Calculate the Retinex output R(x,y,s) at each scale. The formula is as follows:

[0043] R(x,y,s)=log(I optical(x,y))-log(F(x,y,s));

[0044] Among them, R(x,y,s) represents the Retinex output at position (x,y) with scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s;

[0045] S323, introduce an adaptive fusion strategy to dynamically adjust the weights w(s) of different scales according to the image content to obtain the enhanced image R final (x,y)), the formula is as follows:

[0046] R final (x,y)=∑ s w(s)·R(x,y,s);

[0047] Among them, R final (x,y) represents the fused Retinex output at position (x,y), w(s) represents the weight of scale s, and R(x,y,s) represents the Retinex output at position (x,y) at scale s;

[0048] S324, using an improved adaptive weight adjustment method, by introducing adaptive parameters α(x, y) and β(x, y) based on local features of the image, the formula is as follows:

[0049] w(s)=α(x,y)·exp(-β(x,y)·s 2 );

[0050] Wherein, α(x,y) represents the adaptive gain parameter at the position (x,y), β(x,y) represents the adaptive attenuation parameter at the position (x,y), and s represents the scale parameter;

[0051] S325, output R for the fused Retinex final (x,y) is normalized so that its pixel value range is between 0 and 255.

[0052] Optionally, the S33 specifically includes:

[0053] S331, preprocessing the optical image data I optical Perform adaptive Gaussian filtering to reduce noise and obtain the smoothed image I smooth (x,y);

[0054] S332, calculate the gradient of the smoothed image to obtain the gradient amplitude G(x, y) and the gradient direction θ(x, y), the formula is as follows:

[0055]

[0056] Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the smoothed image in the x and y directions respectively;

[0057] S333, perform improved non-maximum suppression on the gradient amplitude G(x, y), refine the edge, and obtain the refined gradient image G nms (x, y), the improvement includes the following formula:

[0058]

[0059] Among them, G nms (x,y) is the gradient amplitude at position (x,y) after non-maximum suppression;

[0060] S334, using a local adaptive double threshold detection method, the refined gradient image G nms (x, y) for edge detection, the formula is as follows:

[0061]

[0062] Among them, E(x,y) represents the edge detection result at the position (x,y), T high (x,y) represents the high threshold at position (x,y), T low (x,y) represents the low threshold at position (x,y). If G nms (x,y) between T low (x,y) and T high If the distance between (x, y) is less than 1, the edge is tracked by the hysteresis threshold method;

[0063] S335, using an adaptive edge connection algorithm, post-processing the detected edge E(x, y), removing isolated weak edge points, and retaining connected strong edges. The formula is as follows:

[0064] E final (x,y)=connect(E(x,y),θ(x,y));

[0065] Among them, E final (x,y) represents the final edge detection result at position (x,y), connect represents the adaptive edge connection algorithm, which dynamically connects edge points according to the edge direction θ(x,y);

[0066] Optionally, the S4 specifically includes:

[0067] S41, from the enhanced optical image data I optical and LiDAR data L LiDAR Extract important terrain features and use adaptive edge detection algorithm to extract the contour of the object E optical (x,y) and elevation data H LiDAR (x,y);

[0068] S42, combined with optical image data I optical and LiDAR data L LiDAR , generate a preliminary three-dimensional digital elevation model, the formula is as follows:

[0069] DEM initial (x,y)=α·I optical (x,y)+β·L LiDAR (x,y);

[0070] Among them, DEM initial (x,y) represents the preliminary digital elevation model value at the position (x,y), I optical (x, y) represents the pixel value of the optical image at position (x, y), L LiDAR (x, y) represents the elevation value of the LiDAR data at the position (x, y), and α and β are the combined weight parameters;

[0071] S43, using a multi-scale fusion strategy, refine the preliminary digital elevation model to generate a high-precision three-dimensional digital elevation model. The formula is as follows:

[0072] DEM refined (x,y)=∑ s w(s)·DEM initial (x,y,s);

[0073] Among them, EM refined (x, y) represents the refined digital elevation model value at the position (x, y), DEM initial (x, y, s) represents the preliminary digital elevation model value at the position (x, y) at scale s, and w(s) is the weight parameter at scale s;

[0074] S44. Generate digital topographic map, combining the refined digital elevation model and terrain contour data E optical (x, y), using adaptive filtering algorithm, the formula is as follows:

[0075] DTM(x,y)=g(DEM refined (x,y),E optical (x,y));

[0076] Among them, DTM (x, y) represents the digital topographic map value at the location (x, y), DEMrefined (x, y) represents the refined digital elevation model value at the position (x, y), E optical (x, y) represents the terrain contour data at the position (x, y), and g represents the adaptive filtering function that combines the refined digital elevation model and the terrain contour to generate a digital terrain map.

[0077] Optionally, the S5 specifically includes:

[0078] S51, perform multi-source data preprocessing on the data from different sensors, firstly, the optical image data I optical 、LiDAR data LiDAR and thermal imaging data T thermal Perform noise removal, geometric correction and radiation correction to obtain the preprocessed data I optical,pre , L LiDAR,pre and T thermal,pre ;

[0079] S52. Use the deep learning-based spatial alignment algorithm to accurately align the preprocessed data to ensure that multi-source data matches in the same geographic coordinate system. The formula is as follows:

[0080] I aligned (x,y)=align(I optical,pre (x,y),L LiDAR,pre (x,y),T thermal,pre (x,y));

[0081] Among them, I aligned (x, y) represents the aligned multi-source data, and align represents the function of spatial alignment using a deep learning model;

[0082] S53. Use a multi-scale convolutional neural network to extract and fuse multi-scale features of the aligned data to generate high-precision comprehensive geographic information data. The formula is as follows:

[0083] I fused (x,y)=MSCNN(I aligned (x,y));

[0084] Among them, I fused (x, y) represents the fused multi-source data, and MSCNN represents the multi-scale convolutional neural network;

[0085] S54, the fused high-precision geographic information data I fused (x, y) is adaptively enhanced, the formula is as follows:

[0086] I enhanced (x,y)=enhance(I fused(x,y));

[0087] Among them, I enhanced (x, y) represents the enhanced geographic information data, and enhance represents the adaptive enhancement function;

[0088] S55, the enhanced high-precision geographic information data I enhanced (x, y) is integrated into the geographic information system to support map making, spatial analysis and various application scenarios. The formula is as follows:

[0089] GIS data =integrate(I enhanced (x,y));

[0090] Among them, GIS data Represents geographic information data integrated into , and integrate represents the data integration function.

[0091] The beneficial effects of the present invention are:

[0092] (1) Significantly improve image resolution and mapping accuracy: UAVs are equipped with optical sensors and thermal imaging sensors, and combined with advanced image processing algorithms (such as contrast-limited adaptive histogram equalization, improved multi-scale Retinex processing, and adaptive edge detection algorithms) to enhance the pre-processed images. Even under complex terrain and lighting conditions, image quality has been significantly improved, and the contours of objects are clearer, thereby greatly improving image resolution and mapping accuracy in complex terrain areas.

[0093] (2) Realize multi-sensor data fusion: The present invention adopts multi-sensor fusion technology to comprehensively process optical image data, LiDAR data and thermal imaging data. Through adaptive fusion strategy and multi-scale fusion algorithm, high-precision three-dimensional digital elevation model and digital terrain map are generated, which improves the integrity and accuracy of terrain data and can provide more detailed and accurate terrain information.

[0094] (3) High-precision 3D terrain modeling: By combining enhanced images and LiDAR data, the present invention generates high-precision 3D digital elevation models and digital terrain maps. By introducing adaptive fusion strategies and multi-scale fusion algorithms, the digital elevation model is further refined, significantly improving the accuracy of 3D terrain modeling and providing a solid data foundation for applications in geographic information systems.

[0095] (4) Efficient data preprocessing and enhancement: The present invention describes in detail the data preprocessing steps, including denoising, geometric correction and radiation correction, to ensure the high quality of the collected data. The preprocessed image is processed using advanced image enhancement algorithms to improve the image contrast, grayscale balance and edge clarity.

[0096] (5) Powerful GIS integration and application capabilities: The present invention integrates high-precision terrain data into the geographic information system, supporting a variety of geographic analysis and application scenarios, such as environmental monitoring, urban planning, and disaster assessment. Through adaptive data integration and enhanced processing technology, the practicality and application value of geographic information system data are improved, making the surveying and mapping results more practical and able to better serve various geographic information application needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0097] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0098] Figure 1 This is a flow chart of a surveying and mapping image enhancement system based on UAV remote sensing technology and its implementation method proposed by the present invention;

[0099] Figure 2 The present invention provides a flow chart of the image enhancement processing steps of a surveying and mapping image enhancement system based on UAV remote sensing technology and its implementation method. DETAILED DESCRIPTION

[0100] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0101] refer to Figure 1 and Figure 2 , a surveying and mapping image enhancement system based on UAV remote sensing technology and its implementation method, comprising the following steps:

[0102] S1. Use the UAV platform to carry optical sensors, LiDAR and thermal imaging sensors to conduct autonomous flight and remote sensing data collection in the target area;

[0103] S2. Preprocessing the collected remote sensing data, including denoising, geometric correction and radiation correction;

[0104] S3, using image enhancement algorithm to perform contrast stretching, histogram equalization and edge detection on the preprocessed image;

[0105] S4, extract important terrain features from the enhanced image and generate high-precision three-dimensional digital elevation model and digital terrain map in combination with LiDAR data;

[0106] S5. Perform multi-source data fusion on data from different sensors and integrate the processed high-precision terrain data into the geographic information system to support map making and spatial analysis.

[0107] In this implementation manner, the S1 specifically includes:

[0108] S11. Select and configure the UAV platform to enable it to carry optical sensors, LiDAR sensors, and thermal imaging sensors;

[0109] S12, setting the flight path and parameters, including flight altitude, speed and route overlap;

[0110] S13, start the UAV, fly autonomously along a predetermined path, and synchronously collect optical images, LiDAR data, and thermal imaging data during the flight;

[0111] S14, the collected optical image data is recorded as I optical , LiDAR data is denoted as L LiDAR , thermal imaging data is recorded as T thermal ;

[0112] S15, real-time monitoring of the UAV flight status and sensor working conditions;

[0113] S16. After the flight is completed, recycle the drone and optical , L LiDAR and T thermal Transmitted to ground processing station.

[0114] In this implementation, S2 specifically includes:

[0115] S21, using median filtering or non-local mean filtering to collect the optical image data I optical Perform denoising processing;

[0116] S22, to I optical Perform geometric correction to correct the image geometric distortion caused by the change of the UAV's flight attitude;

[0117] S23, to I optical Perform radiation correction to correct radiation errors caused by changes in lighting conditions;

[0118] S24, collect the LiDAR data L LiDAR Perform denoising to remove noise from LiDAR data;

[0119] S25, for LLiDAR Perform geometric correction;

[0120] S26. Use appropriate filters to eliminate thermal imaging data T thermal Noise in

[0121] S27, T thermal Perform radiation correction to correct radiation errors in thermal imaging data.

[0122] In this implementation, S3 specifically includes:

[0123] S31, preprocessing the optical image data I optical Perform contrast-limited adaptive histogram equalization;

[0124] S32, to I optical An improved multi-scale Retinex process is performed to balance the grayscale distribution of the image. The improved multi-scale Retinex process formula is as follows:

[0125] R final (x,y)=∑ s w(s)·(log(I optical (x,y))-log(F(x,y,s)));

[0126] Among them, R final (x,y) represents the fused Retinex output at position ((x,y), w(s) represents the weight of scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s;

[0127] S33, using the improved Canny edge detection algorithm, I optical Perform edge detection to extract edge information from the image. The improved Canny edge detection algorithm formula is as follows:

[0128]

[0129] Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the image in the x and y directions respectively;

[0130] S34, the pre-processed LiDAR data L LiDAR Performing improved adaptive filtering processing;

[0131] S35, using improved multi-scale Retinex processing and adaptive edge detection algorithm, the pre-processed thermal imaging data T thermal Perform contrast enhancement and edge detection.

[0132] In this implementation manner, the S32 specifically includes:

[0133] S321, preprocessing the optical image data I optical Use Gaussian filters of different scales for blurring to obtain a multi-scale blurred image F(x, y, s);

[0134] S322. Calculate the Retinex output R(x,y,s) at each scale. The formula is as follows:

[0135] R(x,y,s)=log(I optical (x,y))-log(F(x,y,s));

[0136] Among them, R(x,y,s) represents the Retinex output at position (x,y) with scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s;

[0137] S323, introduce an adaptive fusion strategy to dynamically adjust the weights w(s) of different scales according to the image content to obtain the enhanced image R final (x,y)), the formula is as follows:

[0138] R final (x,y)=∑ s w(s)·R(x,y,s);

[0139] Among them, R final (x,y) represents the fused Retinex output at position (x,y), w(s) represents the weight of scale s, and R(x,y,s) represents the Retinex output at position (x,y) at scale s;

[0140] S324, using an improved adaptive weight adjustment method, by introducing adaptive parameters α(x, y) and β(x, y) based on local features of the image, the formula is as follows:

[0141] w(s)=α(x,y)·exp(-β(x,y)·s 2 );

[0142] Wherein, α(x,y) represents the adaptive gain parameter at the position (x,y), β(x,y) represents the adaptive attenuation parameter at the position (x,y), and s represents the scale parameter;

[0143] S325, output R for the fused Retinex final (x,y) is normalized so that its pixel value range is between 0 and 255.

[0144] In this implementation manner, the S33 specifically includes:

[0145] S331, preprocessing the optical image data I optical Perform adaptive Gaussian filtering to reduce noise and obtain the smoothed image I smooth (x,y);

[0146] S332, calculate the gradient of the smoothed image to obtain the gradient amplitude G(x, y) and the gradient direction θ(x, y), the formula is as follows:

[0147]

[0148] Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the smoothed image in the x and y directions respectively;

[0149] S333, perform improved non-maximum suppression on the gradient amplitude G(x, y), refine the edge, and obtain the refined gradient image G nms (x, y), the improvement includes the following formula:

[0150]

[0151] Among them, G nms (x,y) is the gradient amplitude at position (x,y) after non-maximum suppression;

[0152] S334, using a local adaptive double threshold detection method, the refined gradient image G nms (x, y) for edge detection, the formula is as follows:

[0153]

[0154] Among them, E(x,y) represents the edge detection result at the position (x,y), T high (x,y) represents the high threshold at position (x,y), T low (x,y) represents the low threshold at position (x,y). If G nms(x,y) between T low (x,y) and T high If the distance between (x, y) is less than 1, the edge is tracked by the hysteresis threshold method;

[0155] S335, using an adaptive edge connection algorithm, post-processing the detected edge E(x, y), removing isolated weak edge points, and retaining connected strong edges. The formula is as follows:

[0156] E final (x,y)=connect(E(x,y),θ(x,y));

[0157] Among them, E final (x,y) represents the final edge detection result at position (x,y), connect represents the adaptive edge connection algorithm, which dynamically connects edge points according to the edge direction θ(x,y);

[0158] In this implementation, S4 specifically includes:

[0159] S41, from the enhanced optical image data I optical and LiDAR data L LiDAR Extract important terrain features and use adaptive edge detection algorithm to extract the contour of the object E optical (x,y) and elevation data H LiDAR (x,y);

[0160] S42, combined with optical image data I optical and LiDAR data L LiDAR , generate a preliminary three-dimensional digital elevation model, the formula is as follows:

[0161] DEM initial (x,y)=α·I optical (x,y)+β·L LiDAR (x,y);

[0162] Among them, DEM initial (x,y) represents the preliminary digital elevation model value at the position (x,y), I optical (x, y) represents the pixel value of the optical image at position (x, y), L LiDAR (x, y) represents the elevation value of the LiDAR data at the position (x, y), and α and β are the combined weight parameters;

[0163] S43, using a multi-scale fusion strategy, refine the preliminary digital elevation model to generate a high-precision three-dimensional digital elevation model. The formula is as follows:

[0164] DEM refined (x,y)=∑ sw(s)·DEM initial (x,y,s);

[0165] Among them, EM refined (x, y) represents the refined digital elevation model value at the position (x, y), DEM initial (x, y, s) represents the preliminary digital elevation model value at the position (x, y) at scale s, and w(s) is the weight parameter at scale s;

[0166] S44. Generate digital topographic map, combining the refined digital elevation model and terrain contour data E optical (x, y), using adaptive filtering algorithm, the formula is as follows:

[0167] DTM(x,y)=g(DEM refined (x,y),E optical (x,y));

[0168] Among them, DTM (x, y) represents the digital topographic map value at the location (x, y), DEM refined (x, y) represents the refined digital elevation model value at the position (x, y), E optical (x, y) represents the terrain contour data at the position (x, y), and g represents the adaptive filtering function that combines the refined digital elevation model and the terrain contour to generate a digital terrain map.

[0169] In this implementation manner, S5 specifically includes:

[0170] S51, perform multi-source data preprocessing on the data from different sensors, firstly, the optical image data I optical 、LiDAR data LiDAR and thermal imaging data T thermal Perform noise removal, geometric correction and radiation correction to obtain the preprocessed data I optical,pre , L LiDAR,pre and T thermal,pre ;

[0171] S52. Use the deep learning-based spatial alignment algorithm to accurately align the preprocessed data to ensure that multi-source data matches in the same geographic coordinate system. The formula is as follows:

[0172] I aligned (x,y)=align(I optical,pre (x,y),L LiDAR,pre (x,y),T thermal,pre (x,y));

[0173] Among them, I aligned(x, y) represents the aligned multi-source data, and align represents the function of spatial alignment using a deep learning model;

[0174] S53. Use a multi-scale convolutional neural network to extract and fuse multi-scale features of the aligned data to generate high-precision comprehensive geographic information data. The formula is as follows:

[0175] I fused (x,y)=MSCNN(I aligned (x,y));

[0176] Among them, I fused (x, y) represents the fused multi-source data, and MSCNN represents the multi-scale convolutional neural network;

[0177] S54, the fused high-precision geographic information data I fused (x, y) is adaptively enhanced, the formula is as follows:

[0178] I enhanced (x,y)=enhance(I fused (x,y));

[0179] Among them, I enhanced (x, y) represents the enhanced geographic information data, and enhance represents the adaptive enhancement function;

[0180] S55, the enhanced high-precision geographic information data I enhanced (x, y) is integrated into the geographic information system to support map making, spatial analysis and various application scenarios. The formula is as follows:

[0181] GIS data =integrate(I enhanced (x,y));

[0182] Among them, GIS data Represents geographic information data integrated into , and integrate represents the data integration function.

[0183] This embodiment not only provides a surveying and mapping image enhancement system based on UAV remote sensing technology and its implementation method, but also takes into account the needs of environmental monitoring and disaster assessment, making the assessment more comprehensive and accurate.

[0184] Embodiment 1:

[0185] In a certain mountainous area, the terrain is complex, with clouds and fog all year round and great changes in lighting conditions. Due to its special terrain characteristics, this mountainous area requires accurate mapping and terrain modeling to provide reliable data support for local environmental monitoring, disaster assessment and urban planning. However, traditional mapping methods are difficult to obtain high-precision mapping data under such complex terrain conditions. Therefore, UAV remote sensing technology is used, combined with multi-sensor fusion and advanced image processing algorithms, to carry out detailed mapping and terrain modeling of the area.

[0186] In order to achieve accurate mapping of the mountainous area, we first prepared a drone equipped with optical and thermal imaging sensors. The drone flew autonomously in the target area according to the predetermined flight path, collecting high-resolution optical image data and thermal imaging data. The entire flight process lasted for 3 days, 8 hours a day, and a total of about 5,000 optical images and 2,000 thermal imaging images were collected.

[0187] After data collection was completed, the collected remote sensing data was preprocessed. The median filter and non-local mean filter were used to denoise the optical image, eliminating most of the noise. Geometric correction and radiation correction were performed using high-precision ground control points and standard radiation correction methods to ensure that the image geometry and radiation intensity were consistent with the actual situation.

[0188] The preprocessed image data is contrast enhanced by the contrast-limited adaptive histogram equalization algorithm, making the image details clearer. Subsequently, the image is gray-scale equalized using the improved multi-scale Retinex algorithm, effectively equalizing the brightness distribution of the image. Finally, the edge information in the image is extracted by the adaptive Canny edge detection algorithm, making the contour of the object clearer.

[0189] Combining the enhanced optical image data and thermal imaging data, a high-precision 3D digital elevation model and digital terrain map were generated. In order to further improve the accuracy of the 3D terrain model, we adopted an adaptive fusion strategy and a multi-scale fusion algorithm to refine the initially generated DEM. The final 3D terrain model has a high degree of accuracy and can accurately reflect the terrain characteristics of the mountainous area.

[0190] In order to verify the effectiveness of the present invention, we selected a typical area in the mountainous area for detailed data analysis. The area is about 50 square kilometers, including mountains, forests and some villages. Through UAV remote sensing technology, we collected a large amount of remote sensing data, processed and analyzed the data, and obtained the following results:

[0191] Table 1 Data comparison table of traditional method and method of the present invention

[0192] project Traditional methods Method of the present invention Data collection time 14 days 3 days Image resolution 10 meters 2 meters Surveying accuracy ±5 m ±1m Data Integrity 80％ 95％ Terrain model generation time 7 days 2 days Terrain model accuracy ±10 m ±2 m Application scenario support Environmental monitoring, urban planning Environmental monitoring, urban planning, disaster assessment

[0193] As can be seen from the table, by adopting the method of the present invention, the data acquisition time is greatly shortened from 14 days of the traditional method to 3 days. At the same time, the image resolution is improved from 10 meters of the traditional method to 2 meters, and the mapping accuracy is improved from ±5 meters to ±1 meter. The data integrity is also improved from 80% to 95%, the terrain model generation time is shortened from 7 days to 2 days, and the terrain model accuracy is improved from ±10 meters to ±2 meters. In addition, the method of the present invention has also expanded the application scenarios. In addition to environmental monitoring and urban planning, it can also be used for disaster assessment, etc. These data fully prove that the present invention has significant advantages in image resolution and mapping accuracy in complex terrain areas. Through multi-sensor fusion and advanced image processing algorithms, we have not only improved the quality and accuracy of the data, but also greatly shortened the time for data acquisition and processing, providing reliable data support for various application scenarios.

[0194] In summary, the present invention significantly improves the image resolution and surveying and mapping accuracy in complex terrain areas through the combination of UAV remote sensing technology, multi-sensor fusion and image processing algorithms. It has high efficiency, high precision and broad application prospects, solves many problems of traditional surveying and mapping methods under complex terrain conditions, and provides a solid data foundation for its application in geographic information systems.

[0195] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A surveying and mapping image enhancement method based on UAV remote sensing technology, characterized in that: The following steps are involved: S1. Use the UAV platform to carry optical sensors, LiDAR and thermal imaging sensors to conduct autonomous flight and remote sensing data collection in the target area; S2. Preprocessing the collected remote sensing data, including denoising, geometric correction and radiation correction; S3, using image enhancement algorithm to perform contrast stretching, histogram equalization and edge detection on the preprocessed image; S4, extract important terrain features from the enhanced image and generate high-precision three-dimensional digital elevation model and digital terrain map in combination with LiDAR data; S5. Multi-source data fusion of data from different sensors, and integration of processed high-precision terrain data into geographic information systems to support map making and spatial analysis; The S4 specifically includes: S41, from the enhanced optical image data I optical and LiDAR data L LiDAR Extract important terrain features and use adaptive edge detection algorithm to extract the contour of the object E optical (x,y) and elevation data H LiDAR (x,y); S42, combined with optical image data I optical and LiDAR data L LiDAR , generate a preliminary three-dimensional digital elevation model, the formula is as follows: DEM initial (x,y)=α1·I optical (x,y)+β1·L LiDAR (x,y); Among them, DEM initial (x,y) represents the preliminary digital elevation model value at the position (x,y), I optical (x, y) represents the pixel value of the optical image at position (x, y), L LiDAR (x, y) represents the elevation value of the LiDAR data at the position (x, y), α1 and β1 are the combined weight parameters; S43, using a multi-scale fusion strategy, refine the preliminary digital elevation model to generate a high-precision three-dimensional digital elevation model. The formula is as follows: DEM refined (x,y)=∑ s w(s)·DEM initial (x,y,s); Among them, DEM refined (x, y) represents the refined digital elevation model value at the position (x, y), DEM initial (x, y, s) represents the preliminary digital elevation model value at the position (x, y) at scale s, and w(s) is the weight parameter at scale s; S44. Generate digital topographic map, combining the refined digital elevation model and terrain contour data E optical (x, y), using adaptive filtering algorithm, the formula is as follows: DTM(x,y)=g(DEM refined (x,y),E optical (x,y)); Among them, DTM (x, y) represents the digital topographic map value at the location (x, y), DEM refined (x, y) represents the refined digital elevation model value at the position (x, y), E optical (x, y) represents the contour data of the object at the position (x, y), and g represents the adaptive filtering function that combines the refined digital elevation model and the contour of the object to generate a digital terrain map; The S5 specifically includes: S51, perform multi-source data preprocessing on the data from different sensors, firstly, the optical image data I optical 、LiDAR data LiDAR and thermal imaging data T thermal Perform noise removal, geometric correction and radiation correction to obtain the preprocessed data I optical,pre , L LiDAR,pre and T thermal,pre ; S52. Use the deep learning-based spatial alignment algorithm to accurately align the preprocessed data to ensure that multi-source data matches in the same geographic coordinate system. The formula is as follows: I aligned (x,y)=align(I optical,pre (x,y),L LiDAR,pre (x,y),T thermal,pre (x,y)); Among them, I aligned (x, y) represents the aligned multi-source data, and align represents the function of spatial alignment using a deep learning model; S53. Use a multi-scale convolutional neural network to extract and fuse multi-scale features of the aligned data to generate high-precision comprehensive geographic information data. The formula is as follows: I fused (x,y)=MSCNN(I aligned (x,y)); Among them, I fused (x, y) represents high-precision comprehensive geographic information data, and MSCNN represents multi-scale convolutional neural network; S54. High-precision comprehensive geographic information data I fused (x, y) is adaptively enhanced, the formula is as follows: I enhanced (x,y)=enhance(I fused (x,y)); Among them, I enhanced (x, y) represents the enhanced geographic information data, and enhance represents the adaptive enhancement function; S55, the enhanced geographic information data I enhanced (x, y) is integrated into the geographic information system to support map making and spatial analysis. The formula is as follows: GIS data =integrate(I enhanced (x,y)); Among them, GIS data It represents geographic information data integrated into the geographic information system, and integrate represents the data integration function.

2. The method for enhancing surveying and mapping images based on UAV remote sensing technology according to claim 1, characterized in that: The S1 specifically includes: S11. Select and configure the UAV platform to enable it to carry optical sensors, LiDAR sensors, and thermal imaging sensors; S12, setting the flight path and parameters, including flight altitude, speed and route overlap; S13, start the UAV, fly autonomously along a predetermined path, and synchronously collect optical images, LiDAR data, and thermal imaging data during the flight; S14, the collected optical image data is recorded as I optical , LiDAR data is denoted as L LiDAR , thermal imaging data is recorded as T thermal ; S15, real-time monitoring of the UAV flight status and sensor working conditions; S16. After the flight is completed, recycle the drone and optical , L LiDAR and T thermal Transmitted to ground processing station.

3. The method for enhancing surveying and mapping images based on UAV remote sensing technology according to claim 1, characterized in that: The S2 specifically includes: S21, using median filtering or non-local mean filtering to filter the collected optical image data I optical Perform denoising processing; S22, to I optical Perform geometric correction to correct the image geometric distortion caused by the change of the UAV's flight attitude; S23, to I optical Perform radiation correction to correct radiation errors caused by changes in lighting conditions; S24, collect the LiDAR data L LiDAR Perform denoising to remove noise from LiDAR data; S25, for L LiDAR Perform geometric correction; S26, using a filter to eliminate thermal imaging data T thermal Noise in S27, T thermal Perform radiation correction to correct radiation errors in thermal imaging data.

4. The method for enhancing surveying and mapping images based on UAV remote sensing technology according to claim 1, characterized in that: The S3 specifically includes: S31, preprocessing the optical image data I optical Perform contrast-limited adaptive histogram equalization; S32, to I optical An improved multi-scale Retinex process is performed to balance the grayscale distribution of the image. The improved multi-scale Retinex process formula is as follows: R final (x,y)=∑ s w(s)·(log(I optical (x,y))-log(F(x,y,s))); Among them, R final (x,y) represents the fused Retinex output at position (x,y), w(s) represents the weight of scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s; S33, using the improved Canny edge detection algorithm, I optical Perform edge detection to extract edge information from the image. The improved Canny edge detection algorithm formula is as follows: Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the image in the x and y directions respectively; S34, the pre-processed LiDAR data L LiDAR Performing improved adaptive filtering processing; S35, using improved multi-scale Retinex processing and adaptive edge detection algorithm, the pre-processed thermal imaging data T thermal Perform contrast enhancement and edge detection.

5. The method for enhancing surveying and mapping images based on UAV remote sensing technology according to claim 4, characterized in that: The S32 specifically includes: S321, preprocessing the optical image data I optical Use Gaussian filters of different scales for blurring to obtain a multi-scale blurred image F(x, y, s); S322. Calculate the Retinex output R(x,y,s) at each scale. The formula is as follows: R(x,y,s)=log(I optical (x,y))-log(F(x,y,s)); Among them, R(x,y,s) represents the Retinex output at position (x,y) with scale s, I optical (x,y) represents the pixel value of the optical image at position (x,y), and F(x,y,s) represents the pixel value of the blurred image at position (x,y) with a scale of s; S323, introduce an adaptive fusion strategy to dynamically adjust the weights w(s) of different scales according to the image content to obtain the enhanced image R final (x,y)), the formula is as follows: R final (x,y)=∑ s w(s)·R(x,y,s); Among them, R final (x,y) represents the fused Retinex output at position (x,y), w(s) represents the weight of scale s, and R(x,y,s) represents the Retinex output at position (x,y) at scale s; S324, using an improved adaptive weight adjustment method, by introducing adaptive parameters α(x, y) and β(x, y) based on local features of the image, the formula is as follows: w(s)=α(x,y)·exp(-β(x,y)·s 2 ); Wherein, α(x,y) represents the adaptive gain parameter at the position (x,y), β(x,y) represents the adaptive attenuation parameter at the position (x,y), and s represents the scale parameter; S325, output R for the fused Retinex final (x,y) is normalized so that its pixel value range is between 0 and 255.

6. The method for enhancing surveying and mapping images based on UAV remote sensing technology according to claim 4, characterized in that: The S33 specifically includes: S331, preprocessing the optical image data I optical Perform adaptive Gaussian filtering to reduce noise and obtain the smoothed image I smooth (x,y); S332, calculate the gradient of the smoothed image to obtain the gradient amplitude G(x, y) and the gradient direction θ(x, y), the formula is as follows: Among them, G(x,y) represents the gradient amplitude at the position (x,y), θ(x,y) represents the gradient direction at the position (x,y), and Represent the gradient of the smoothed image in the x and y directions respectively; S333, perform improved non-maximum suppression on the gradient amplitude G(x, y), refine the edge, and obtain the refined gradient image G nms (x, y), the improvement includes the following formula: Among them, G nms (x,y) is the gradient amplitude at position (x,y) after non-maximum suppression; S334, using a local adaptive double threshold detection method, the refined gradient image G nms (x, y) for edge detection, the formula is as follows: Among them, E(x,y) represents the edge detection result at the position (x,y), T high (x,y) represents the high threshold at position (x,y), T low (x,y) represents the low threshold at position (x,y). If G nms (x,y) between T low (x,y) and T high If the distance between (x, y) is less than 1, the edge is tracked by the hysteresis threshold method; S335, using an adaptive edge connection algorithm, post-processing the detected edge E(x, y), removing isolated weak edge points, and retaining connected strong edges. The formula is as follows: E final (x,y)=connect(E(x,y),θ(x,y)); Among them, E final (x,y) represents the final edge detection result at the position (x,y), and connect represents the adaptive edge connection algorithm, which dynamically connects the edge points according to the gradient direction θ(x,y).

Images