Unmanned aerial vehicle inspection method based on VR holographic imaging

Abstract

The invention relates to the technical field of virtual reality, in particular to an unmanned aerial vehicle inspection method based on VR holographic imaging. The invention provides an unmanned aerial vehicle inspection method based on VR holographic imaging, and the method is characterized in that pixel points of a multi-scale digital image are extracted, a wavelet transform coefficient of the multi-scale digital image after denoising is obtained through maximum likelihood estimation, and the wavelet transform coefficient is used for carrying out noise recognition on the multi-scale digital image to obtain variance distribution of noise signals of the multi-scale digital image; constructing a multi-scale image denoising model to perform fine processing on a multi-scale digital image to obtain a denoised multi-scale digital image, obtaining an image enhancement gradient field according to the multi-scale digital holographic imaging model, combining wavelet decomposition and introducing weighted fusion to obtain a high-frequency detail component of the multi-scale digital image; and obtaining multi-scale digital holographic imaging according to an image fusion rule. The invention aims to solve the problem that the brightness and resolution of images collected and shot by the unmanned aerial vehicle are reduced.

Description

Technical Field

[0001] This invention relates to the field of virtual reality technology, and in particular to a drone inspection method based on VR holographic imaging. Background Technology

[0002] As an advanced type of aircraft, drones have demonstrated their wide application value in military, civilian, and scientific research fields in recent years. Compared with human labor, drones have advantages such as wide coverage, fast inspection speed, and low cost, significantly improving efficiency and safety.

[0003] When performing missions, drones need to fly in varied and complex environments, such as mountains, dense forests, buildings, and severe weather conditions. Factors such as the structural parameters of photoelectric sensors, light intensity, and transmission direction can affect the brightness and resolution of images collected and captured by drones during inspections, making it difficult to achieve the desired results.

[0004] In summary, there is an urgent need for a drone inspection method based on VR holographic imaging to improve the problems of low brightness and low resolution in the photos captured during existing drone inspection processes.

[0005] Holographic imaging technology, also known as virtual imaging technology or phantom imaging, is a technology that uses the principles of light interference and diffraction to record and reproduce the true three-dimensional image of an object. By filtering out zero-order diffraction spots and reducing speckle noise, it improves the visual effect and the brightness of the image. Summary of the Invention

[0006] The purpose of this invention is to improve the problem of reduced brightness and resolution of images collected and captured by UAVs in existing technologies by filtering out zero-order diffraction spots and reducing speckle noise.

[0007] The present invention provides a drone inspection method based on VR holographic imaging, comprising the following steps:

[0008] The saliency of pixels in multi-scale images is calculated by defining a probability function for the distribution of pixels in images captured by a drone.

[0009] The saliency of the multi-scale image pixels is combined with the distribution variance of the multi-scale digital image pixels to extract the multi-scale digital image pixels.

[0010] Maximum likelihood estimation is used to estimate the wavelet transform coefficients of denoised multi-scale digital images and to identify the variance distribution of noise signals in multi-scale digital images.

[0011] A multi-scale image denoising model is constructed based on the variance distribution of the multi-scale digital image noise signal to refine the multi-scale digital image and obtain the denoised multi-scale digital image.

[0012] Construct a multi-scale digital holographic imaging model and obtain the image enhancement gradient field based on the multi-scale digital holographic imaging model;

[0013] The image enhancement gradient field combines wavelet decomposition and introduces weighted fusion of high-frequency detail components of multi-scale digital images;

[0014] The high-frequency detail components of the multi-scale digital image are used for multi-scale digital holographic imaging according to image fusion rules.

[0015] The present invention provides a drone inspection method based on VR holographic imaging, which has the technical advantages of appropriate image brightness, clear images, and ease of implementation.

[0016] Optionally, the exponent of the probability function is the quotient of the distance from the digital image pixel to the center point and the square of the digital image parameters, and the probability function is used to distinguish between background boundary images and VR scene images.

[0017] Optionally, when calculating the saliency of multi-scale image pixels, the distance between pixels and the contrast are inversely proportional. The saliency of multi-scale image pixels is calculated by utilizing the spatial function of the multi-scale digital image and the differences between pixel values ​​in the digital image.

[0018] Optionally, when extracting multi-scale digital image pixels, prior knowledge is used to measure the weight of multi-scale digital images within a salient region, and human visual habits are combined to treat the pixels of the digital image and the central position of the VR scene as the area weight of the salient region.

[0019] Optionally, when calculating the wavelet transform coefficients after denoising the multi-scale digital image, the noise variance of the multi-scale digital image is estimated. Since the neighborhood variance of the multi-scale digital image has a relatively strong correlation, the approximate value of the wavelet coefficients can be obtained by maximum likelihood estimation.

[0020] Optionally, during noise identification, the noise coefficient of the digital image is the sum of the wavelet coefficients of the digital image and the noise locations in the digital image.

[0021] Optionally, the step of refining the multi-scale digital image to obtain the denoised multi-scale digital image includes:

[0022] The quadratic composite wavelet transform technique is used to achieve comprehensive acquisition of wavelet parameters of multi-scale digital images;

[0023] Based on the superposition metric of quantum state collapse, the variance of wavelet coefficients of multi-scale digital images is estimated;

[0024] The wavelet factors of the denoised multi-scale digital image are obtained, and then the wavelet correlation parameters are used to perform operations on the acquired multi-scale digital image.

[0025] The denoised multi-scale digital image is obtained by inverse wavelet transform.

[0026] Optionally, in the process of constructing the multi-scale digital holographic imaging model, it is necessary to combine the pixel coordinates of the multi-scale digital image, the thickness of the multi-scale digital holographic imaging, the length of the multi-scale digital holographic imaging, and the imaging time difference.

[0027] Optionally, the step of obtaining the image enhancement gradient field includes:

[0028] Magnify the results of multi-scale digital holographic imaging;

[0029] The image enhancement gradient field is obtained by combining the image enhancement processing value with the multi-scale digital image enhancement function.

[0030] Optionally, the wavelet decomposition takes the multi-scale digital image detection matrix as a starting point and uses the image enhancement gradient field to process the multi-scale digital holographic image. Attached Figure Description

[0031] Figure 1 This is a structural diagram of the device for implementing a UAV inspection method based on VR holographic imaging provided by the present invention;

[0032] Figure 2 This is a comparison chart of the imaging clarity of the UAV inspection method based on VR holographic imaging provided by this invention with other methods. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0034] This invention provides a drone inspection method based on VR holographic imaging, comprising the following steps:

[0035] S1. Calculate the saliency of pixels in multi-scale images by defining a probability function for the distribution of pixels in images captured by a drone.

[0036] S2. The saliency of the multi-scale image pixels is combined with the distribution variance of the multi-scale digital image pixels to extract the multi-scale digital image pixels.

[0037] S3. The maximum likelihood estimation is used to estimate the wavelet transform coefficients of the denoised multi-scale digital image and to perform noise identification on the multi-scale digital image, which is based on the variance distribution of the noise signal of the multi-scale digital image.

[0038] S4. Construct a multi-scale image denoising model based on the variance distribution of the multi-scale digital image noise signal to refine the multi-scale digital image and obtain the denoised multi-scale digital image.

[0039] S5. Construct a multi-scale digital holographic imaging model and obtain the image enhancement gradient field based on the multi-scale digital holographic imaging model;

[0040] S6. The image enhancement gradient field is combined with wavelet decomposition and weighted fusion is introduced to the high-frequency detail components of the multi-scale digital image;

[0041] S7. The high-frequency detail components of the multi-scale digital image are used for multi-scale digital holographic imaging according to image fusion rules.

[0042] The structural diagram of the device for implementing a UAV inspection method based on VR holographic imaging provided in this embodiment is shown below. Figure 1 As shown:

[0043] A multi-scale digital holographic imaging system was built under the Windows 8 operating system. The system's host frequency is 5.0GHz, memory is 16GB, resolution is 300M, image sensor size is 1 / 3.2 inches, and camera pixel accuracy is 0.075mm.

[0044] When performing step S1, the exponent of the probability function is the quotient of the distance from the digital image pixel to the center point and the square of the digital image parameters. The probability function is used to distinguish between background boundary images and VR scene images.

[0045] To distinguish between background boundary images and VR scene images, a probability function P(x) is defined, where x represents the feature vector of a pixel or a group of pixels in the image. This function aims to quantify the probability that a pixel or pixel region belongs to the background boundary or the VR scene.

[0046] This embodiment of the invention uses Bayes' theorem to define this probability function.

[0047] Suppose there are two categories: C1 represents the background boundary image, and C2 represents the VR scene image. We can define the following probability function:

[0048] P(C1∣x)=P(x)P(x∣C1)P(C1)

[0049] P(C2∣x)=P(x)P(x∣C2)P(C2)

[0050] Where P(C1|x) and P(C2|x) are the probabilities that a pixel belongs to the background boundary image and the VR scene image, respectively, given the feature vector x.

[0051] P(x|C1) and P(x|C2) are the conditional probabilities of feature vector x appearing in the background boundary image and the VR scene image, respectively. These can be estimated using the training dataset.

[0052] P(C1) and P(C2) are the prior probabilities of the background boundary image and the VR scene image in the total dataset, respectively.

[0053] P(x) is the total probability of the occurrence of the eigenvector x, and is usually used as a normalization factor.

[0054] In practical applications, since directly calculating P(x) can be difficult, it is usually better to compare P(C1|).

[0055] The relative magnitudes of P(C2|x) and P(C2|x) can be determined by comparing their numerators (i.e., likelihood multiplied by prior probability).

[0056] Furthermore, to simplify computation and improve efficiency, simpler feature representations and classifiers, such as Support Vector Machines (SVM), Decision Trees, Random Forests, or Deep Neural Networks, might be used. These classifiers can directly output the probability that a pixel or pixel region belongs to a certain category.

[0057] When calculating the saliency of pixels in a multi-scale image, the distance between pixels and the contrast are inversely proportional. The saliency of pixels in a multi-scale image is calculated by utilizing the spatial function of the multi-scale digital image and the differences between pixel values ​​in the digital image.

[0058] First, the spatial function of the multi-scale digital image needs to be calculated. Next, this spatial function is multiplied by the differences between the pixel values ​​of the image, and these multiplications are summed to determine the saliency of the digital image pixels.

[0059] A spatial function is a function that describes the spatial relationships between pixels in an image. In multi-scale digital images, since the image is decomposed into representations at different scales, the spatial function also needs to consider information from these different scales. Specifically, a spatial function can be constructed by calculating the spatial distance or similarity between each pixel and its neighboring pixels. These spatial distances or similarities can be based on different metrics, such as Euclidean distance, Manhattan distance, or cosine similarity.

[0060] To assess the saliency of pixel values ​​in an image, we need to calculate the difference between each pixel and its surrounding pixels. This difference can be obtained by calculating the absolute difference, squared difference, or other metrics between pixel values. When calculating the difference, multi-scale information also needs to be considered, as image representations at different scales may contain different saliency information.

[0061] After obtaining the spatial function and pixel value differences, we combine these two factors to calculate the saliency of a pixel. Specifically, we can multiply the spatial function and pixel value differences, and then sum these multiplications. In this way, the saliency of each pixel can be determined by both its spatial relationship with surrounding pixels and its pixel value differences. The summation process ensures that we obtain a global saliency measure, rather than relying solely on local information.

[0062] When performing step S2, when extracting multi-scale digital image pixels, the weights of multi-scale digital images in salient regions are assigned by using prior knowledge and indicators to measure the weights of the salient regions. Combined with human visual habits, the pixels of the digital image and the central position of the VR scene are used as the area weights of the salient regions.

[0063] In the process of in-depth comparison and analysis of globally salient regions, a comprehensive research methodology was developed by integrating global and local salient features from multi-scale digital images. During this process, a significant feature was observed in a virtual reality (VR) environment: locally salient regions exhibited high-density pixel clusters, and these pixels displayed a high degree of similarity. Compared to globally salient regions, the boundaries of locally salient regions were more blurred, often blending into the surrounding scattered background, increasing the complexity of their direct identification.

[0064] To achieve accurate quantification of the saliency of multi-scale digital images, a series of scientific evaluation metrics were designed and introduced based on profound prior knowledge. These metrics were used to meticulously evaluate multi-scale digital images within salient regions, and appropriate weights were assigned to them according to the evaluation results. Subsequently, these weights were deeply integrated with the distribution characteristics of pixels, thereby constructing an efficient multi-scale digital image quantization system.

[0065] Furthermore, the perceptual characteristics of the human visual system are fully considered. When initially observing an image, human vision tends to focus first on clustered regions, and the edges of these clustered regions are usually less salient. Based on this understanding, the relationship between the pixels of the digital image and the center position of the VR scene is used as an important factor to determine the area weight of salient regions. This innovative approach not only improves the accuracy of saliency detection but also provides strong support for subsequent image processing and visual analysis.

[0066] Based on the above, multi-scale digital image pixels are extracted:

[0067] First, the interval S of each seed point is calculated using the total number of pixels N in the image and the number of superpixels K to be extracted. Then, the feature vector V of each seed point in the image is extracted sequentially using the interval S. The feature vector V typically includes color values ​​and image coordinates, i.e.: V = [l, a, b, x, y]

[0068] Where [l,a,b] represents the color value of the seed point in the CIE-Lab color space, and [x,y] is the image coordinate of each seed point.

[0069] Next, the K-means clustering algorithm is used to cluster the feature vectors of each pixel. During the clustering process, the difference D between pixels can be calculated using the following formula:

[0070] D = dc + (m / S)ds

[0071] Where dc represents the distance between pixels in the CIE-Lab color space, ds represents the spatial distance between pixels, m is the weighting parameter that balances color and spatial distance, and S is the spacing between seed points.

[0072] By calculating the difference D between each pixel and each seed point, the pixel can be assigned to the cluster to which the nearest seed point belongs, thus completing the extraction of superpixels.

[0073] When performing step S3, when calculating the wavelet transform coefficients after denoising the multi-scale digital image, the noise variance of the multi-scale digital image is estimated. Since the neighborhood variance of the multi-scale digital image has a relatively strong correlation, the approximate value of the wavelet coefficients can be obtained by maximum likelihood estimation.

[0074] In general, for a specific range M, we explore effective methods for predicting noise variance in multi-scale digital images. During this process, significant correlations are found between the neighborhood variances of multi-scale digital images.

[0075] Specifically, when pixel (i, j) is within the range M, the maximum likelihood estimation method in statistics is used to conduct an in-depth analysis and calculation of the noise variance of multi-scale digital images. This method successfully obtains an approximate value for the noise variance, which has high accuracy and reliability.

[0076] The wavelet coefficients of the denoised multi-scale digital image are calculated using maximum likelihood estimation (ML) and approximate values ​​of the variance of multi-scale digital image noise:

[0077] First, perform a multi-scale wavelet transform on the original image to obtain wavelet coefficients at different scales. This is typically done using libraries such as PyWavelets.

[0078] import pywt

[0079] import numpy as np

[0080] #Load the image (assuming it's a grayscale image)

[0081] image = np.load('image.npy') # Replace with the actual image loading code

[0082] # Perform wavelet transform

[0083] coeffs = pywt.dwt2(image, 'haar') # Use the Haar wavelet; other wavelets can be selected as needed.

[0084] cA,(cH,cV,cD)=coeffs #cA is the approximation coefficient, cH is the horizontal detail coefficient, cV is the vertical detail coefficient, and cD is the diagonal detail coefficient.

[0085] Next, the noise variance at each scale is approximated using maximum likelihood estimation or other methods. This is typically based on the statistical properties of the wavelet coefficients:

[0086] Assuming the noise primarily exists in the detail coefficients, we can calculate the variance of these coefficients as an approximation of the noise variance.

[0087] def estimate_noise_variance(coeffs,scale_factor=1.0):

[0088] #Extract detail coefficients and flatten

[0089] details=np.concatenate([cH.flatten(),cV.flatten(),cD.flatten()])

[0090] # Calculate the variance (Note: The scale_factor is used here to adjust the variance estimate, which may need to be adjusted according to the actual situation)

[0091] noise_variance=np.var(details)*scale_factor

[0092] return noise variance

[0093] #Calculate noise variance

[0094] noise_variance=estimate_noise_variance(coeffs)

[0095] print(f"Estimated noise variance:{noise_variance}")

[0096] Given the noise variance, maximum likelihood estimation or other Bayesian methods can be used to denoise wavelet coefficients.

[0097] # Define a simple soft threshold function

[0098] def soft_thresholding(coeffs,threshold):

[0099] sign = np.sign(coeffs)

[0100] return sign*np.maximum(np.abs(coeffs)-threshold,0)

[0101] # Calculate the threshold based on the noise variance (e.g., using the general threshold formula: threshold = sqrt(2 * log(N)) * sigma, where N is the number of coefficients and sigma is the noise standard deviation).

[0102] threshold=np.sqrt(2*np.log(len(coeffs[1][0].flatten())))*np.sqrt(noise_variance)

[0103] # Apply soft thresholding to detail factor

[0104] cH_denoised=soft_thresholding(cH,threshold)

[0105] cV_denoised=soft_thresholding(cV,threshold)

[0106] cD_denoised=soft_thresholding(cD,threshold)

[0107] #Keep the approximation coefficients unchanged

[0108] cA_denoised=cA

[0109] #Reconstructed and denoised wavelet coefficients

[0110] coeffs_denoised=(cA_denoised,(cH_denoised,cV_denoised,cD_denoised))

[0111] The process requires attention to the following: Different wavelet bases have a significant impact on denoising performance and should be selected based on image characteristics and denoising requirements. The choice of threshold is crucial to the denoising effect and can be adjusted based on factors such as noise variance and image size. The above process can serve as a starting point for iterative optimization; further improvements in denoising performance can be achieved by adjusting parameters and introducing more complex models or algorithms.

[0112] When performing step S3, during noise identification, the noise coefficient of the digital image is the sum of the wavelet coefficients of the digital image and the noise location of the digital image.

[0113] When performing step S4, the step of refining the multi-scale digital image to obtain the denoised multi-scale digital image includes:

[0114] The quadratic composite wavelet transform technique is used to achieve comprehensive acquisition of wavelet parameters of multi-scale digital images;

[0115] Based on the superposition metric of quantum state collapse, the variance of wavelet coefficients of multi-scale digital images is estimated;

[0116] The wavelet factors of the denoised multi-scale digital image are obtained, and then the wavelet correlation parameters are used to perform operations on the acquired multi-scale digital image.

[0117] The denoised multi-scale digital image is obtained by inverse wavelet transform.

[0118] When performing step S5, the process of constructing the multi-scale digital holographic imaging model requires combining the pixel coordinates of the multi-scale digital image, the thickness of the multi-scale digital holographic imaging, the length of the multi-scale digital holographic imaging, and the imaging time difference.

[0119] The object beam and the reference beam can be represented as complex wave fields U_o(x,y) and Ur(x,y), respectively. The hologram H(x,y) is the interference result of these two light waves:

[0120] H(x,y)=|U_o(x,y)+U_r(x,y)|^2

[0121] This can be further elaborated as follows:

[0122] H(x,y)=|U_o(x,y)|^2+|U_r(x,y)|^2+2Re{U_o(x,y)U_r^*(x,y)}

[0123] Where Re{} denotes taking the real part, and U_r^*(x,y) is the conjugate complex number of the reference light wave field.

[0124] The goal of numerical reconstruction is to recover the object's light wave field U_o(x,y) (or its complex amplitude distribution) from the hologram H(x,y). This typically involves numerical methods for diffraction calculations, such as Fresnel transform and convolution algorithms.

[0125] Taking the convolution algorithm as an example, the reconstruction process can be represented as:

[0126] U_o'(x,y,z)=F^(-1){F{H(x,y)}×H_sys(f_x,f_y,z)}

[0127] Where F{} and F^(-1){} represent the Fourier transform and inverse Fourier transform, respectively, and H_sys(f_x,f_y,z) is the system transfer function, which describes the diffraction process of the light wave field from the object plane to the reconstruction plane. z is the reconstruction distance.

[0128] When performing step S5, the step of obtaining the image enhancement gradient field includes:

[0129] Magnify the results of multi-scale digital holographic imaging;

[0130] The image enhancement gradient field is obtained by combining the image enhancement processing value with the multi-scale digital image enhancement function.

[0131] Magnification:

[0132] Iscaled = ScaleUp(Ihologram)

[0133] Image enhancement processing:

[0134] Evalue = Enhance(Iscaled)

[0135] Calculate the image enhancement gradient field:

[0136] Gfield = Efunction(Evalue)

[0137] Gfield=GradientField(Evalue)

[0138] ScaleUp is a scaling function that can enlarge the holographic image to the desired size.

[0139] Enhance is an image enhancement function that can process a magnified image according to a specific algorithm or model to extract or enhance useful information in the image.

[0140] GradientField is a function that calculates the image enhancement gradient field based on the enhancement values.

[0141] When performing step S6, the wavelet decomposition takes the multi-scale digital image detection matrix as the starting point and uses the image enhancement gradient field to process the multi-scale digital holographic image.

[0142] First, wavelet decomposition is performed on the multi-scale digital holographic imaging results. Wavelet decomposition can decompose the image into sub-bands of different scales and directions, including low-frequency components (approximation coefficients) and high-frequency components (detail coefficients).

[0143] Let Ihologram be the original result of multi-scale digital holographic imaging. Wavelet decomposition can be expressed as:

[0144] {Aj,{Dj,k}k}=WaveletDecompose(Ihologram)

[0145] Where Aj is the low-frequency component (approximation coefficient) of the j-th layer, {Dj,k}k is the set of high-frequency components (detail coefficients) of the j-th layer, and k represents different directions (such as horizontal, vertical, diagonal, etc.).

[0146] Next, the high-frequency components obtained from the decomposition are subjected to weighted fusion. The purpose of weighted fusion is to combine high-frequency detail components at different scales and directions according to specific weight assignments to obtain enhanced high-frequency detail information.

[0147] Let Wj,k be the weights in the j-th layer and k-th direction. The weighted fusion can be expressed as:

[0148] Dfused=j,k∑Wj,k·Dj,k

[0149] Among them, Dfused is the high-frequency detail component after weighted fusion.

[0150] Ultimately, the weighted fused high-frequency detail component Dfused can be used as the high-frequency detail component for multi-scale digital holographic imaging, and can be used for subsequent image processing or VR scene rendering.

[0151] Combining the above steps, we can obtain a simplified formula framework:

[0152] Dfused=j,k∑Wj,k·(WaveletDecompose(Ihologram))j,khigh-freq

[0153] Where (WaveletDecompose(Ihologram))j,khigh-freq represents the high-frequency component in the j-th layer and k-th direction after wavelet decomposition.

[0154] The choice of weights Wj,k should be determined based on the specific application scenario and requirements, and may involve trade-offs in image quality assessment, detail preservation, computational complexity, and other aspects.

[0155] Based on the high-frequency detail components of the image, the fusion rules for various regions in the multi-scale digital holographic imaging process are given:

[0156] The high-frequency detail components are weighted and averaged according to preset weights, which can balance information from different scales or sources:

[0157] Dfused(x,y)=i∑wi·Di(x,y)

[0158] Where wi is the weight of the i-th high-frequency detail component, and ∑iwi=1.

[0159] Multi-scale digital holographic imaging is achieved using image fusion rules, as shown below:

[0160] The fused high-frequency detail components are combined with low-frequency components (such as a combination of one or all layers in Aj) to reconstruct the final multi-scale digital holographic imaging result.

[0161] Let Ifused be the reconstructed image, then:

[0162] Ifused=ReconstructImage(Aselected,Dfused)

[0163] Here, Aselected is the selected low-frequency component (which may be an approximation coefficient of a certain layer or a combination of multiple approximation coefficients), and ReconstructImage is an image reconstruction function that reconstructs the final image based on the low-frequency component and the high-frequency detail component.

[0164] The imaging clarity comparison diagram between the VR holographic imaging-based UAV inspection method provided in this embodiment and other methods is shown in the figure below. Figure 2 As shown:

[0165] The method provided by this invention achieves an imaging clarity of over 90%, which improves the clarity of images captured during UAV inspections compared to imaging methods based on compressed sensing algorithms and dynamic speckle interferometry.

[0166] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A drone inspection method based on VR holographic imaging, characterized in that, Includes the following steps: The saliency of pixels in multi-scale images is calculated by defining a probability function for the distribution of pixels in images captured by a drone. The saliency of the multi-scale image pixels is combined with the distribution variance of the multi-scale digital image pixels to extract the multi-scale digital image pixels. Maximum likelihood estimation is used to estimate the wavelet transform coefficients of denoised multi-scale digital images and to identify the variance distribution of noise signals in multi-scale digital images. A multi-scale image denoising model is constructed based on the variance distribution of the multi-scale digital image noise signal to refine the multi-scale digital image and obtain the denoised multi-scale digital image. Construct a multi-scale digital holographic imaging model and obtain the image enhancement gradient field based on the multi-scale digital holographic imaging model; The image enhancement gradient field combines wavelet decomposition and introduces weighted fusion of high-frequency detail components of multi-scale digital images; The high-frequency detail components of the multi-scale digital image are used for multi-scale digital holographic imaging according to image fusion rules.

2. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... The exponent of the probability function is the quotient of the distance from the digital image pixel to the center point and the square of the digital image parameters. The probability function is used to distinguish between background boundary images and VR scene images.

3. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... When calculating the saliency of pixels in a multi-scale image, the distance between pixels and the contrast are inversely proportional. The saliency of pixels in a multi-scale image is calculated by utilizing the spatial function of the multi-scale digital image and the differences between pixel values ​​in the digital image.

4. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... When extracting multi-scale digital image pixels, prior knowledge is used to measure the weight of multi-scale digital images in salient regions, and combined with human visual habits, the pixels of the digital image and the central position of the VR scene are regarded as the area weight of the salient region.

5. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... When calculating the wavelet transform coefficients after denoising a multi-scale digital image, the noise variance of the multi-scale digital image is estimated. The neighborhood variance of the multi-scale digital image has a strong correlation, and the approximate value of the wavelet coefficients can be obtained through maximum likelihood estimation.

6. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... During noise identification, the noise coefficient of the digital image is the sum of the wavelet coefficients of the digital image and the noise location in the digital image.

7. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... The steps for refining the multi-scale digital image to obtain the denoised multi-scale digital image include: The quadratic composite wavelet transform technique is used to achieve comprehensive acquisition of wavelet parameters of multi-scale digital images; Based on the superposition metric of quantum state collapse, the variance of wavelet coefficients of multi-scale digital images is estimated; The wavelet factors of the denoised multi-scale digital image are obtained, and then the wavelet correlation parameters are used to perform operations on the acquired multi-scale digital image. The denoised multi-scale digital image is obtained by inverse wavelet transform.

8. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... In the process of constructing the multi-scale digital holographic imaging model, it is necessary to combine the pixel coordinates of the multi-scale digital image, the thickness of the multi-scale digital holographic imaging, the length of the multi-scale digital holographic imaging, and the imaging time difference.

9. The UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... The steps for obtaining the image enhancement gradient field include: Magnify the results of multi-scale digital holographic imaging; The image enhancement gradient field is obtained by combining the image enhancement processing value with the multi-scale digital image enhancement function.

10. A UAV inspection method based on VR holographic imaging according to claim 1, characterized in that... The wavelet decomposition starts with the multi-scale digital image detection matrix and uses the image enhancement gradient field to process the multi-scale digital holographic image.

Images