Hyperspectral remote sensing image target detection method and device, readable storage medium and electronic equipment
By using pre-constructed filters and constrained energy minimization algorithms in hyperspectral remote sensing images, the background object signal is suppressed, and the problem of unsatisfactory large object detection effect in hyperspectral remote sensing images is solved, and higher detection accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202411499654.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-25
- Publication Date
- 2025-02-25
AI Technical Summary
In hyperspectral remote sensing images, when the signal distribution of the target objects is dense, it is easy to affect the signal of the background objects, resulting in unsatisfactory large-target detection results.
By acquiring three-dimensional hyperspectral remote sensing image data, the data is processed using a pre-constructed hyperspectral image filter, the target spectral data set and sample autocorrelation matrix are determined, and combined with the constraint energy minimization algorithm, a hyperspectral image filter is constructed to suppress background object signals.
It effectively overcomes the impact of the target spectrum and background spectrum on the autocorrelation matrix, improves the detection accuracy and accuracy of large targets, and reduces the false detection rate.
Smart Images
Figure CN119516358A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of image detection technology, and in particular to a method, device, readable storage medium and electronic device for detecting targets in hyperspectral remote sensing images. Background Art
[0002] Hyperspectral images are three-dimensional images that combine imaging technology and spectral technology to form a unified map. Each pixel in a hyperspectral remote sensing image is a continuous spectrum composed of hundreds or even thousands of bands. The spectral bands usually span visible light, near infrared and mid-infrared parts, ranging from 0.4 microns to 2.5 microns. Each type of ground object corresponds to a spectral curve, and different ground objects correspond to different spectral curves. These image data record the electromagnetic wave signals and energy of the ground objects in detail. Compared with other types of remote sensing images, hyperspectral remote sensing images contain richer spectral information of ground objects and can accurately reflect the subtle spectral differences of ground objects. When processing hyperspectral remote sensing data, according to different task requirements, in order to highlight the characteristics of certain ground objects, by selecting certain specific spectral bands within the spectral range, it can not only provide more basic analysis data of ground objects for the hyperspectral remote sensing image processing algorithm, but also accurately process and analyze the ground objects.
[0003] Among target detection methods, detection methods based on linear mixed models have been widely used. Typically, Harsanyi and Chang proposed a sub-pixel target detector based on a finite impulse response filter, namely the constrained energy minimization method (CEM), which determines a filter vector by analyzing the hyperspectral remote sensing image data and the known target spectral information, and processes the image through the filter vector to obtain the target detection result. The filter vector is used to remove the spectral information of the background objects in the hyperspectral remote sensing image, so that the spectral information of the target objects can pass smoothly, while suppressing the output of the spectral information of the background objects. When the target objects are distributed in a small number and scattered in the hyperspectral remote sensing image, the impact on the signal of the background objects can be ignored, that is, when the constrained energy minimization algorithm detects small targets, its accuracy is better, but when the signal of the target object is densely distributed in the hyperspectral image and occupies a large proportion in the image, it can have a great impact on the signal of the background objects, that is, the detection effect on large targets is not ideal, and the constrained energy minimization algorithm is not suitable for the detection of large targets. Summary of the invention
[0004] In view of this, the embodiments of the present application propose a hyperspectral remote sensing image target detection method, device, readable storage medium and electronic device, aiming to solve the problem that when the signal of the target object is densely distributed in the hyperspectral image and occupies a large proportion in the image, it can have a great impact on the signal of the background object, that is, the detection effect of the large target is not ideal.
[0005] To achieve the above-mentioned purpose, an embodiment of the present application provides a hyperspectral remote sensing image target detection method, comprising: obtaining three-dimensional hyperspectral remote sensing image data to be processed; processing the three-dimensional hyperspectral remote sensing image data using a pre-constructed hyperspectral image filter to obtain target image data; wherein the process of pre-constructing a hyperspectral image filter comprises: obtaining pre-processed three-dimensional hyperspectral remote sensing image data; determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set, and outputting the target image data.
[0006] Optionally, before obtaining the preprocessed three-dimensional hyperspectral remote sensing image data, the hyperspectral remote sensing image target detection method includes: using a uniform area method, a geostatistical method, a local mean or a local standard deviation method to evaluate data noise in the three-dimensional hyperspectral remote sensing image data and remove data noise; using a dark target atmospheric correction algorithm to remove atmospheric absorption window bands in the three-dimensional hyperspectral remote sensing image data.
[0007] Optionally, determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set, comprises: obtaining a sample autocorrelation matrix, wherein the expression of the sample autocorrelation matrix is:
[0008]
[0009] Among them, R is the weighted value of the autocorrelation matrix of all pixels in the three-dimensional hyperspectral remote sensing image data; c is a constant, g(x) is a monotonically increasing function, f(x, y) is a vector similarity judgment function, d is the target spectral dataset, r i is the preprocessed data pixel, and N is the total number of pixels.
[0010] Optionally, determining the hyperspectral image filter based on the sample autocorrelation matrix and the target spectral dataset includes: acquiring pixel data and a target spectral dataset vector in the preprocessed three-dimensional hyperspectral remote sensing image data; and determining the hyperspectral image filter based on an inverse matrix of the sample autocorrelation matrix of the pixel data and a target spectral dataset vector.
[0011] Optionally, the hyperspectral image filter is determined based on the inverse matrix of the sample autocorrelation matrix of the pixel data and the target spectral data set vector, including: obtaining an average energy expression of the filter output signal; determining a linear constraint optimization function based on the converted average energy expression; determining the hyperspectral image filter to be optimized according to the linear constraint optimization function; obtaining an optimization objective function of the hyperspectral image filter to be optimized, wherein the optimization objective function is the ratio of the average energy expression of the target spectral data set to the average energy expression of the background object signal; iteratively inputting the preprocessed three-dimensional hyperspectral remote sensing image data and the target spectral data set into the hyperspectral image filter to be optimized to obtain the hyperspectral image filter when the optimization objective function reaches the maximum.
[0012] Optionally, the expression of the hyperspectral image filter is:
[0013]
[0014] Among them, R * represents the autocorrelation matrix of three-dimensional hyperspectral remote sensing image data, d is the target spectral dataset, and ω is the hyperspectral image filter.
[0015] Optionally, the expression of the optimization objective function is, including:
[0016]
[0017] Where W represents the hyperspectral image filter, T represents the autocorrelation matrix of the target spectral data set, E represents the average energy of the target spectral data set after passing through the filter, R represents the autocorrelation matrix of the three-dimensional hyperspectral remote sensing image data to be processed, and F represents the average output energy of all the three-dimensional hyperspectral remote sensing image data to be processed in the image.
[0018] In order to solve the above problems, the present application also provides a hyperspectral remote sensing image target detection device, including: an acquisition module, used to acquire three-dimensional hyperspectral remote sensing image data to be processed; a processing module, used to process the three-dimensional hyperspectral remote sensing image data using a pre-constructed hyperspectral image filter to obtain target image data; wherein the process of pre-constructing a hyperspectral image filter includes: acquiring pre-processed three-dimensional hyperspectral remote sensing image data; determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; determining the hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set, and an output module, outputting the target image data.
[0019] In order to solve the above problems, the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the hyperspectral remote sensing image target detection method provided in any of the aforementioned embodiments can be implemented.
[0020] In order to solve the above problems, the present application also provides a terminal / electronic device / server, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the hyperspectral remote sensing image target detection method provided in any of the aforementioned embodiments.
[0021] The embodiments of the present application propose a method, device, readable storage medium and electronic device for detecting targets in hyperspectral remote sensing images, which obtain three-dimensional hyperspectral remote sensing image data to be processed; process the three-dimensional hyperspectral remote sensing image data using a pre-constructed hyperspectral image filter to obtain target image data; wherein the process of pre-constructing the hyperspectral image filter includes: obtaining pre-processed three-dimensional hyperspectral remote sensing image data; determining a target spectrum data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectrum data set; determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectrum data set, combining the autocorrelation matrix in the constrained energy minimization (CEM) algorithm with the target spectrum set to determine a hyperspectral image filter, and reducing the influence of the target object on the matrix when participating in the operation as a background object through the signal suppression factor of the background object, effectively overcoming the influence of the target spectrum and the background spectrum on the autocorrelation matrix, and being able to achieve better suppression of scattered backgrounds compared to traditional algorithms, and effectively improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 is a flow chart of a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0023] Figure 2 Houston image data in the hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0024] Figure 3 A spectral curve of a target building in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0025] Figure 4 A target prior distribution map in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0026] Figure 5 The target detection results before and after improvement in the hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0027] Figure 6 The ROC curve in the hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0028] Figure 7 A hyperspectral data image in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0029] Figure 8 A spectral curve of a target vehicle in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0030] Fig. 9 A target prior distribution map in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0031] FIG10 is a comparison of detection results before and after improvement in a hyperspectral remote sensing image target detection method provided in one embodiment of the present application;
[0032] Fig.11 The ROC curve of the hyperspectral remote sensing image target detection method provided in one embodiment of the present application is
[0033] Fig.12 A schematic diagram of the structure of a hyperspectral remote sensing image target detection device provided in one embodiment of the present application;
[0034] Fig.13 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0035] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be limited in any way, and the various embodiments can be combined and referenced with each other without contradiction.
[0036] An embodiment of the present application proposes a hyperspectral remote sensing image target detection method, which is applied to an electronic device, wherein the electronic device can be a terminal or a server. This embodiment and the following embodiments are all described by taking the server as an example. The implementation details of the hyperspectral remote sensing image target detection method proposed in this embodiment are described in detail below. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.
[0037] Figure 1 is a flowchart of a hyperspectral remote sensing image target detection method provided in one embodiment of the present application, with reference to Figure 1 , a hyperspectral remote sensing image target detection method, comprising:
[0038] S10, obtaining three-dimensional hyperspectral remote sensing image data to be processed;
[0039] Among them, three-dimensional hyperspectral remote sensing image data refers to image data with three-dimensional characteristics obtained through hyperspectral remote sensing technology. The three dimensions are:
[0040] Spatial dimension: Like traditional two-dimensional images, hyperspectral images also have width and height, that is, the spatial resolution in the x and y directions, which are expressed as rows and columns of the image.
[0041] Spectral dimension: This is what makes hyperspectral images unique. They capture continuous, narrow-band spectral information at each spatial pixel. This spectral information constitutes the third dimension of the image, namely spectral resolution, which is expressed as the number of bands in the image.
[0042] Radiation dimension: records the intensity of radiation reflected or emitted from the ground, usually combined with the spectral dimension to form a spectral curve for each pixel.
[0043] This multidimensional nature of hyperspectral remote sensing image data enables it to provide richer information than traditional remote sensing images. In a hyperspectral image, each pixel has a unique spectral signature that can be used to identify and classify the material composition and object type of the surface. For example, hyperspectral images can distinguish different types of vegetation, rocks, soil, and other surface features.
[0044] Before step S10, the hyperspectral remote sensing image target detection method further includes:
[0045] S31. Use a uniform area method, a geostatistical method, a local mean or a local standard deviation method to evaluate and remove data noise in the three-dimensional hyperspectral remote sensing image data, and use a dark target atmospheric correction algorithm to remove atmospheric absorption window bands in the three-dimensional hyperspectral remote sensing image data.
[0046] S20. Processing the three-dimensional hyperspectral remote sensing image data using a pre-constructed hyperspectral image filter to obtain target image data; wherein the process of pre-constructing the hyperspectral image filter comprises: acquiring pre-processed three-dimensional hyperspectral remote sensing image data; determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; and determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set.
[0047] It should be noted that the target image data is the target detection result of the hyperspectral remote sensing image of the present invention.
[0048] Before using the pre-built hyperspectral image filter to process the three-dimensional hyperspectral remote sensing image data, it is necessary to pre-build the hyperspectral image filter. The process of pre-building the hyperspectral image filter includes:
[0049] S21. Select a target spectrum from the three-dimensional hyperspectral remote sensing image data to obtain a target spectrum data set and a vector thereof.
[0050] S22. Determine a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set.
[0051] Autocorrelation matrix It can be regarded as all pixels r in the hyperspectral image i (i=1,2,...,N) autocorrelation matrix is weighted by weights. R can suppress background information. At the same time, we hope that the influence of the target signal d on R is minimized or even eliminated, and the influence of the target signal of no interest on R is as large as possible. Therefore, the sample autocorrelation matrix R is introduced here. * .
[0052] The expression of the sample autocorrelation matrix is:
[0053]
[0054] Among them, R is the weighted value of the autocorrelation matrix of all pixels in the three-dimensional hyperspectral remote sensing image data; c is a constant, g(x) is a monotonically increasing function, f(x, y) is a vector similarity judgment function, d is the target spectral dataset, r i is the preprocessed data pixel, N is the total number of pixels, the higher the similarity between x and y, the smaller the value of f(x, y); the lower the similarity between x and y, the larger the value of f(x, y).
[0055] S23, determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set.
[0056] S231, obtaining pixel data and target spectral data set vector in the preprocessed three-dimensional hyperspectral remote sensing image data;
[0057] S232: Determine a hyperspectral image filter based on the inverse matrix of the sample autocorrelation matrix of the pixel data and the target spectral data set vector.
[0058] In an embodiment of the present application, step S232 may include the following specific execution steps:
[0059] S2321, obtaining an average energy expression of the filter output signal;
[0060] Among them, the average energy expression of the filter output signal can be:
[0061] like Figure 1 As shown, the hyperspectral image is represented by r. Given the target spectral data set d, it is assumed that there is a FIR filter that can ensure that the target spectral data set d can be output stably, and can ensure that the average energy of the final overall output is minimized, while suppressing the expression of background object information. The filter can be represented by ω. R represents the autocorrelation matrix of the sample, and the average energy expression of the output signal can be:
[0062]
[0063] S2322, constructing a linear constraint optimization function based on the converted average energy expression;
[0064] Specifically, the constrained energy minimization (CEM) algorithm is transformed into a linear constrained optimization problem.
[0065] The linear constrained optimization problem is expressed as:
[0066]
[0067] Among them, d t ω represents the product of the transpose of the target spectral dataset d and the constrained energy minimization filter to be 1, min{ω t Rω} indicates that the average energy of the output signal is minimized.
[0068] S2323. Determine a hyperspectral image filter to be optimized according to the linear constraint optimization function.
[0069] Specifically, the expression of the linear constrained optimization problem is solved to obtain the hyperspectral image filter.
[0070] Among them, the expression of the hyperspectral image filter is:
[0071]
[0072] Among them, R * represents the autocorrelation matrix of three-dimensional hyperspectral remote sensing image data, d is the target spectral dataset, and ω is the hyperspectral image filter.
[0073] S2323, obtaining an optimization objective function of the hyperspectral image filter to be optimized, wherein the optimization objective function is a ratio of an average energy expression of a target spectral data set to an average energy expression of a background object signal;
[0074] By applying the detection operator of the hyperspectral image filter to each pixel of the hyperspectral image, the specific distribution of the target spectral data set d in the remote sensing image can be obtained, thereby achieving the effect of target detection. The constrained energy minimization (CEM) algorithm has been widely used in the detection of small targets and has achieved very good detection results, but in the detection process of large targets, the effect is not ideal and there are great shortcomings. Therefore, the following improvements to the constrained energy minimization algorithm are proposed here.
[0075] In the process of hyperspectral image target detection, in order to highlight the spectral information of the target object, a common method is to suppress the spectral information of the background objects and enhance the spectral vector of the target of interest, so that the spectral information of the target object is more obvious. The constrained energy minimization algorithm mainly relies on the sample autocorrelation matrix R for target detection. The autocorrelation matrix R can effectively suppress the background information, but the target spectral vector d will affect the autocorrelation matrix R.
[0076] Assume D = (d1, d2, ..., d n ) is the target spectral dataset of interest, d i =(d i1 ,d i2 ,...,d im ) T , (i=1,2,...,n) is a single target spectrum vector of interest, W=(w1,w2,...,w m ) is an m-dimensional linear filter.
[0077] Let e i =d t i W,f i =r t iW, i = 1, 2, ..., l. The filter W can not only enhance the target spectral data set d, but also suppress the signal of the background objects to the maximum extent. Therefore, it is necessary to consider how to balance the relationship between the two. The preprocessed three-dimensional hyperspectral remote sensing image data and the target spectral data set are iteratively input into the hyperspectral image filter to be optimized to obtain the maximum value of the optimization objective function λ, where the expression of the optimization objective function λ is as follows:
[0078]
[0079] Where W represents the hyperspectral image filter, T represents the autocorrelation matrix of the target spectral data set, E represents the average energy of the target spectral data set after passing through the filter, R represents the autocorrelation matrix of the three-dimensional hyperspectral remote sensing image data to be processed, and F represents the average output energy of all the three-dimensional hyperspectral remote sensing image data to be processed in the image.
[0080] The constrained energy minimization algorithm only needs to know the information of the target object of interest, and does not need to know the signal of the background object. Therefore, the signal of the background object here is replaced by the output energy of all the three-dimensional hyperspectral remote sensing image data to be processed. According to the expression of the optimization objective function λ, the following formula can be obtained:
[0081] R -1 Tω=λω
[0082] Will Substituting the above formula into the following formula:
[0083]
[0084] When the target spectral data set D contains only one target vector of interest, the autocorrelation matrix of the target sample of interest is T = dd T ,formula
[0085]
[0086] After simplification, we can get:
[0087]
[0088] At this point, by comparing the above formula with the CEM operator, it is found that the above formula has an additional proportional coefficient compared to the CEM operator, that is, the CEM operator is a special case of the above formula.
[0089] In practice, the background signal energy is the average output energy obtained by filtering the signal energy of all ground object pixels in the hyperspectral image, in which the target spectral dataset d not only participates in the calculation of the target spectral dataset energy, but also participates in the calculation of the background signal energy. When the target ground objects are relatively few and scattered in the hyperspectral remote sensing image, the impact on the signal of the background ground objects can be ignored, which also explains why the constrained energy minimization algorithm has better accuracy when detecting small targets. However, when the target spectral dataset d is densely distributed in the hyperspectral image and occupies a large proportion in the image, it can have a great impact on the signal of the background ground objects, so the constrained energy minimization algorithm is not suitable for the detection of large targets.
[0090] S30: Output the target image data.
[0091] Example 1
[0092] Object detection based on Houstonu data
[0093] The hyperspectral data used in this experiment was made public by the laboratory of Dr. Saurabh Prasad at the University of Houston. The data was acquired by NCALM on the University of Houston campus and its vicinity between 16:31 and 18:18 GMT on February 16, 2017. This data is multi-sensor data, including 48-band hyperspectral data, 3-band LiDAR data, and ultra-high resolution images. The data set contains 20 types of ground objects. Figure 2 , as shown in Table 1. The spectral curve of the building for target detection is shown in Figure 3 The distribution of targets in hyperspectral images is shown in Figure 4 shown.
[0094] Table 1 HoustonU data information
[0095]
[0096]
[0097] By analyzing the spectral curve characteristics of the target object, and based on the target feature information, the constrained energy minimization algorithm is used to detect the target before and after improvement, and the following results are obtained: Figure 5 As shown, Figure 5 (a) represents the test results before improvement, Figure 5 (b) shows the improved detection results.
[0098] By comparing the experimental results, it can be clearly seen that some scattered and erroneous targets before the improvement have been greatly improved after the improvement. In order to further make an objective evaluation of the performance of the target detection algorithm before and after the improvement, the above results are quantitatively analyzed using the ROC curve, as shown in the figure below: Figure 6 shown.
[0099] Among them, the blue curve represents the effect of target detection before the algorithm is improved, and its area AUC = 0.73. The yellow curve represents the effect of target detection after the algorithm is improved, and its area AUC = 0.76. The comparison of the ROC curve and its area shows that the performance advantage of the improved algorithm is greatly improved compared with that before the algorithm was improved.
[0100] Example 2
[0101] Object detection based on The Cooke City Dataset
[0102] The hyperspectral dataset used in this experiment was obtained by Snyder et al. using the hymap airborne hyperspectral imaging sensor to conduct reconnaissance in Cook City, Montana, USA. The hyperspectral imager measures the spectral bands in the wavelength range of 0.45 to 2.48 microns. The dataset contains 800 scan lines, each with 280 pixels, and each pixel contains 126 bands. Figure 7 As shown, the spectral curve of the selected car is as follows Figure 8 The distribution of targets in hyperspectral images is shown in Fig. 9 shown.
[0103] By analyzing the spectral curve characteristics of the target object and based on the target feature information, the target is detected before and after improvement using the constrained energy minimization algorithm, and the following results are obtained, as shown in Figure 10, where Figure 10(a) shows the detection result before improvement, and Figure 10(b) shows the detection result after improvement.
[0104] By comparing the experimental results, it can be clearly seen that some scattered and erroneous targets before the improvement have been greatly improved after the improvement. In order to further make an objective evaluation of the performance of the target detection algorithm before and after the improvement, the above results are quantitatively analyzed using the ROC curve, as shown in the figure below: Fig.11 shown.
[0105] Among them, the yellow curve represents the effect of target detection before the algorithm is improved, and its area AUC = 0.79. The blue curve represents the effect of target detection after the algorithm is improved, and its area AUC = 0.80. The comparison of the ROC curve and its area shows that the improved algorithm has no negative effect on the small target detection performance.
[0106] In summary, the constrained energy minimization algorithm, after optimization and improvement, has made great improvements in the accuracy and precision of large target detection. Specifically, the optimized and improved constrained energy minimization algorithm performs better in terms of accuracy. The improved algorithm can better identify targets and can effectively reduce the false detection rate, thereby improving accuracy. At the same time, the improved algorithm has also made significant improvements in false alarm rate. At the same time, while improving the defects of large target detection in the constrained energy minimization algorithm, the effect of the algorithm on small target detection has not been reduced. In short, by optimizing and improving the constrained energy minimization algorithm, the accuracy and precision of large target detection can be effectively improved, thereby further promoting the development of the target detection field.
[0107] The step division of the above methods is only for clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this application; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this application.
[0108] Based on the above embodiments, the present invention also provides a hyperspectral remote sensing image target detection device, referring to Fig.12 The hyperspectral remote sensing image target detection device 100 includes an acquisition module 101, a processing module 102 and an output module 103, wherein the acquisition module 101 is used to acquire the three-dimensional hyperspectral remote sensing image data to be processed; the processing module 102 is used to process the three-dimensional hyperspectral remote sensing image data using a pre-constructed hyperspectral image filter to obtain target image data; the process of pre-constructing the hyperspectral image filter includes: acquiring the pre-processed three-dimensional hyperspectral remote sensing image data; determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set, and the output module 103 is used to output the target image data.
[0109] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.
[0110] Based on the above embodiment, another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the above method embodiment is implemented.
[0111] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM (Read-Only Memory), RAM (Random Access Memory), disk or optical disk and other media that can store program codes.
[0112] Another embodiment of the present application provides an electronic device, whose specific structure can be as follows: Fig.13 As shown, it includes: at least one processor 601; and a memory 602 that is communicatively connected to the at least one processor 601; wherein the memory 602 stores instructions that can be executed by the at least one processor 601, and the instructions are executed by the at least one processor 601 so that the at least one processor 601 can execute the hyperspectral remote sensing image target detection method in the above-mentioned method embodiments.
[0113] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripheral devices, voltage regulators and power management circuits together, which are all well known in the art, so this article will not describe them further. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, and further, the antenna also receives the data and transmits the data to the processor.
[0114] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.
[0115] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.
Claims
1. A hyperspectral remote sensing image target detection method, characterized in that: include: Acquire three-dimensional hyperspectral remote sensing image data to be processed; Processing the three-dimensional hyperspectral remote sensing image data using a pre-built hyperspectral image filter to obtain target image data; The process of pre-building a hyperspectral image filter includes: Obtain preprocessed three-dimensional hyperspectral remote sensing image data; Determine a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determine a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral dataset; The target image data is output.
2. The hyperspectral remote sensing image target detection method according to claim 1, characterized in that: Before obtaining the preprocessed three-dimensional hyperspectral remote sensing image data, the hyperspectral remote sensing image target detection method includes: Using a uniform area method, a geostatistical method, a local mean or a local standard deviation method to evaluate data noise in the three-dimensional hyperspectral remote sensing image data and remove the data noise; The dark target atmospheric correction algorithm is used to remove the atmospheric absorption window band in three-dimensional hyperspectral remote sensing image data.
3. The hyperspectral remote sensing image target detection method according to claim 2, characterized in that: Determining a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determining a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set, including: Obtain a sample autocorrelation matrix, the expression of which is: Among them, R * is the weighted value of the autocorrelation matrix of all pixels in the three-dimensional hyperspectral remote sensing image data; c is a constant, g(x) is a monotonically increasing function, f(x, y) is a vector similarity judgment function, d is the target spectral dataset, r i is the preprocessed data pixel, and N is the total number of pixels.
4. The hyperspectral remote sensing image target detection method according to claim 3, characterized in that: The step of determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral data set comprises: Obtain pixel data and target spectral data set vector in preprocessed three-dimensional hyperspectral remote sensing image data; A hyperspectral image filter is determined based on an inverse matrix of a sample autocorrelation matrix of the pixel data and a target spectral data set vector.
5. The hyperspectral remote sensing image target detection method according to claim 4, characterized in that: The method of determining a hyperspectral image filter based on an inverse matrix of a sample autocorrelation matrix of the pixel data and a target spectral data set vector comprises: Get the average energy expression of the filter output signal; Determine a linear constraint optimization function based on the converted average energy expression; Determining a hyperspectral image filter to be optimized according to the linear constraint optimization function; Obtaining an optimization objective function of the hyperspectral image filter to be optimized, wherein the optimization objective function is a ratio of an average energy expression of a target spectral data set to an average energy expression of a background object signal; The preprocessed three-dimensional hyperspectral remote sensing image data and the target spectral data set are iteratively input into the hyperspectral image filter to be optimized to obtain the hyperspectral image filter when the optimization objective function reaches the maximum.
6. The hyperspectral remote sensing image target detection method according to claim 2, characterized in that: The expression of the hyperspectral image filter is: Among them, R * is the weighted value of the autocorrelation matrix of all pixels in the three-dimensional hyperspectral remote sensing image data, d is the target spectral dataset, and ω is the hyperspectral image filter.
7. The hyperspectral remote sensing image target detection method according to claim 5, characterized in that: The expression of the optimization objective function is: Where W represents the hyperspectral image filter, T represents the autocorrelation matrix of the target spectral data set, E represents the average energy of the target spectral data set after passing through the hyperspectral image filter, R represents the autocorrelation matrix of the three-dimensional hyperspectral remote sensing image data to be processed, and F represents the average output energy of all the three-dimensional hyperspectral remote sensing image data to be processed in the image.
8. A hyperspectral remote sensing image target detection device, characterized in that: include: An acquisition module, used for acquiring three-dimensional hyperspectral remote sensing image data to be processed; A processing module, used for processing the three-dimensional hyperspectral remote sensing image data using a pre-built hyperspectral image filter to obtain target image data; The process of pre-building a hyperspectral image filter includes: Obtain preprocessed three-dimensional hyperspectral remote sensing image data; Determine a target spectral data set based on the three-dimensional hyperspectral remote sensing image data, and determine a sample autocorrelation matrix in a constrained energy minimization algorithm based on the target spectral data set; determining a hyperspectral image filter based on the sample autocorrelation matrix and the target spectral dataset; An output module is used to output the target image data.
9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the hyperspectral remote sensing image target detection method according to any one of claims 1 to 7 can be implemented.
10. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the hyperspectral remote sensing image target detection method as described in any one of claims 1 to 7.
Citation Information
Cited By
Aquaculture net cage detection method based on adaptive variable parameter dynamic learning network
CN119992110A