Data acquisition and processing method and system for photoelectric measurement

Abstract

The invention relates to the technical field of photoelectric measurement, and discloses a data acquisition and processing method and system for photoelectric measurement. The method comprises the steps of obtaining photoelectric data of a target area, performing preprocessing to realize multi-source data space-time registration, and extracting a space-time fusion photoelectric signal; spectral features are obtained according to the signal, a preliminary detection area is divided, and fine target segmentation and feature optimization are completed; constructing a feature extraction model based on time-frequency analysis to perform target classification, and generating an acquisition result; energy distribution changes are obtained through the multi-period data, and dynamic calibration parameters are generated. The system comprises a memory and a processor, and executes corresponding programs to realize the steps. The method solves the problems of multi-source data fusion, target identification, dynamic calibration and the like in the prior art, improves the precision and reliability of photoelectric measurement, and is suitable for photoelectric measurement scenes in multiple fields.

Description

Technical Field

[0001] The present invention relates to the technical field of photoelectric measurement, and in particular to a data acquisition and processing method and system for photoelectric measurement. Background Art

[0002] In the course of modern scientific and technological development, photoelectric measurement technology, thanks to its high precision, non-contact nature, and fast response, has been widely used in numerous fields, including industrial inspection, environmental monitoring, and aerospace. However, the current data acquisition and processing process for photoelectric measurement still faces numerous challenges, severely restricting the further development and application of this technology.

[0003] Photoelectric measurement often requires the acquisition of multi-source data, including data from photoelectric sensor arrays and mobile measurement devices. These different types of data differ in time and space, making them difficult to directly fuse and process. Traditional data acquisition methods typically employ a fixed sensor layout and a single data acquisition mode, which cannot fully adapt to complex and changing measurement environments, resulting in poor temporal and spatial consistency of the data. For example, when measuring in field environments with complex terrain, traditional fixed sensor arrays struggle to fully cover the target area. Data acquired by mobile measurement devices also struggle to match sensor array data due to a lack of effective calibration. This results in significant errors in the data fusion process, impacting the accuracy of the measurement results.

[0004] In the data processing phase, existing data processing algorithms have shortcomings in spectral feature extraction and target segmentation. Traditional spectral feature extraction methods are sensitive to noise and interference in the signal, making it difficult to accurately extract the key spectral features of the measured object, resulting in low target recognition accuracy. During the target segmentation process, segmentation methods based on simple thresholds or fixed rules cannot adapt to the diversity and complexity of different targets, and are prone to over-segmentation or under-segmentation problems, making it impossible to achieve precise target segmentation. For example, when measuring small targets or targets with similar spectral characteristics, traditional methods have difficulty accurately distinguishing the target from the background, reducing the reliability of the measurement.

[0005] Furthermore, existing data acquisition and processing systems lack dynamic calibration mechanisms. During the measurement process, environmental factors (such as changes in light intensity and temperature fluctuations) and variations in device performance can cause drift and deviation in measurement data. However, traditional systems are unable to timely calibrate data based on actual measurement conditions, leading to the accumulation of measurement errors and difficulty ensuring long-term measurement accuracy and stability. This limits the application of photoelectric measurement technology in scenarios such as long-term monitoring.

[0006] In summary, the existing photoelectric measurement data acquisition and processing technology has obvious defects in multi-source data fusion, spectral feature extraction, target segmentation and dynamic calibration. A new technical solution is urgently needed to solve these problems in order to improve the accuracy, reliability and adaptability of photoelectric measurement. Summary of the Invention

[0007] The object of the present invention is to provide a data acquisition and processing method and system for photoelectric measurement to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: a data acquisition and processing method for photoelectric measurement, the method comprising: Acquire optoelectronic data of the target area, including optoelectronic sensor array data and mobile measurement device data; preprocess the acquired optoelectronic data and implement spatiotemporal registration of multi-source data to extract spatiotemporal fusion optoelectronic signals of the target area; Acquiring spectral characteristics of the measurement object based on the spatiotemporal fusion photoelectric signal, dividing a preliminary detection area using the spectral characteristics, intercepting the photoelectric signal of the detection area to perform pattern recognition to complete fine target segmentation, and optimizing the spectral characteristics; A feature extraction model is constructed based on the time-frequency analysis method. The optimized spectral features are used as the model input. The target is classified through multi-scale decomposition. The photoelectric measurement data acquisition results of the target area are generated based on the classification results and the corresponding spectral features. The energy distribution change corresponding to the fine target segmentation result is obtained through the multi-time period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change.

[0009] Preferably, the acquired photoelectric data is preprocessed and multi-source data spatiotemporal registration is implemented to extract the spatiotemporal fusion photoelectric signal of the target area, specifically: Aligning the photoelectric data by spatial grids and time series, obtaining the spatiotemporal correlation of the photoelectric data corresponding to the grids or sequences, and fusing the grids or sequences whose correlations meet preset conditions; Obtaining the phase error during the photoelectric data fusion process, judging the fusion accuracy of the grid or sequence based on the phase error, and performing the next grid fusion when the phase error is within a preset tolerance range until all grids or sequences are fused; The fused photoelectric signal is subjected to baseline correction using multi-scale decomposition, and a dynamic threshold is set according to the signal amplitude fluctuation to suppress noise. The corrected photoelectric signal is obtained, and the background interference component is eliminated and normalized. Obtaining the preprocessed photoelectric signal, extracting local frequency domain features through an adaptive filtering network, performing coherence comparison on the local frequency domain features, presetting a number of feature points, and obtaining the point where the local frequency domain feature with the highest coherence is located as a reference point based on the number of feature points; Using the preprocessed photoelectric data as a reference signal and a target signal, mapping the fiducial point features in the reference signal and the target signal to obtain spatiotemporal correlation features, and matching the spatiotemporal correlation features with local frequency domain features; A matching weight matrix is ​​obtained, and signal registration of the photoelectric data is performed according to the weight matrix, and the spatiotemporal fusion photoelectric signal of the registered target area is extracted.

[0010] Preferably, the spectral characteristics of the measurement object are obtained according to the spatiotemporal fusion photoelectric signal, and the preliminary detection area is divided using the spectral characteristics, specifically: Acquire the spatiotemporal fusion photoelectric signal of the target area, generate the corresponding spectrum image after time-frequency transformation, and perform frequency band segmentation based on the spectrum image to obtain the main frequency component and energy distribution characteristics; Performing regional interception on the spectrum image according to the energy distribution characteristics to suppress high-frequency noise, obtaining the target core frequency band by intercepting the optimized spectrum image, obtaining the time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band, and converting the time domain distribution into a grayscale spectrum; performing sub-pixel frequency domain positioning in the grayscale spectrum to obtain frequency band edge information, iteratively optimizing the frequency band through least squares ellipse fitting based on the edge information, obtaining the energy center of the optimized ellipse, determining the frequency band width based on the energy center, and generating frequency domain features; Generate spectral features based on the main frequency component, energy distribution characteristics and frequency domain characteristics, use historical spatiotemporal fusion photoelectric signals to extract corresponding spectral features to construct a training data set, and train a convolutional neural network with the training data set to segment the measurement target; Each spectral feature is generated based on sliding window regression, and the corresponding feature area is labeled in the frequency domain to achieve preliminary division of the detection area.

[0011] Preferably, the photoelectric signal of the detection area is intercepted and pattern recognition is performed to complete fine target segmentation, and the spectral characteristics are optimized, specifically: Acquire a preliminary detection area, perform clustering in different frequency domain intervals of the preliminary detection area according to energy density thresholds, and select interval peak points in each frequency domain interval; The energy density threshold is set by extracting the distance between the target optimization ellipse energy centers through the spectral characteristics corresponding to the target, and the distance between the peak point and the nearest frequency domain signal is calculated from high frequency to low frequency to determine whether the distance is less than the energy density threshold; If it is less than, the energy distribution information of the frequency domain interval is updated; if it is greater than, the distance difference is obtained; if the distance difference is less than the preset tolerance, the energy distribution information of the frequency domain interval is updated; The boundary range of the detection area is updated according to the energy distribution information of the frequency domain interval, the clustering of all signals is completed through iteration, the fine target segmentation is completed according to the boundary range of the updated detection area, and the spectral characteristics are optimized.

[0012] Preferably, a feature extraction model is constructed based on a time-frequency analysis method, specifically: A feature extraction model is constructed based on wavelet transform to obtain optimized spectral features, which are then imported into the feature extraction model. The time-frequency features of different frequency domain intervals are extracted using convolution kernels. Hidden nodes of corresponding layers are set according to the time-frequency features of different frequency domain intervals, and fusion is performed through a fully connected layer. A weight distribution layer is set in the feature extraction model, and the attention mechanism is used to obtain the dynamic weights of the time-frequency features in different frequency domain intervals to represent the contribution of the time-frequency features; The weighted time-frequency features are compressed again through convolution and average pooling operations, and the compressed time-frequency features are dimensionality reduced and fused before being imported into the fully connected layer. The probability distribution of the target category is calculated by a normalized exponential function, the target classification result is determined according to the probability distribution, and the target classification result of the target area is combined with the corresponding spectral characteristics to generate the photoelectric measurement data acquisition result of the target area.

[0013] Preferably, the energy distribution change corresponding to the fine target segmentation result is obtained by multi-period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change, specifically: Extracting optimized main frequency components and energy distribution characteristics according to the optimized spectral characteristics, obtaining calibration reference points using the optimized main frequency components and energy distribution characteristics, and generating a calibration reference point set; Obtaining a fine target segmentation result through multi-period photoelectric measurement data of the target area, and performing data interpolation on the calibration reference point set according to the fine target segmentation result; The data change trend of the calibration reference point set is obtained to extract the energy distribution change, and the energy distribution change is compared with a preset calibration range. When the energy distribution change exceeds the preset calibration range, a dynamic calibration parameter is generated.

[0014] Preferably, the present invention further includes a data acquisition and processing system for photoelectric measurement, the system comprising: a memory and a processor, wherein the memory includes a data acquisition and processing program for photoelectric measurement, and when the data acquisition and processing program is executed by the processor, the following steps are implemented: Acquire optoelectronic data of the target area, including optoelectronic sensor array data and mobile measurement device data; preprocess the acquired optoelectronic data and implement spatiotemporal registration of multi-source data to extract spatiotemporal fusion optoelectronic signals of the target area; Acquiring spectral characteristics of the measurement object based on the spatiotemporal fusion photoelectric signal, dividing a preliminary detection area using the spectral characteristics, intercepting the photoelectric signal of the detection area to perform pattern recognition to complete fine target segmentation, and optimizing the spectral characteristics; A feature extraction model is constructed based on the time-frequency analysis method. The optimized spectral features are used as the model input. The target is classified through multi-scale decomposition. The photoelectric measurement data acquisition results of the target area are generated based on the classification results and the corresponding spectral features. The energy distribution change corresponding to the fine target segmentation result is obtained through the multi-time period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change.

[0015] Preferably, the spectral characteristics of the measurement object are obtained according to the spatiotemporal fusion photoelectric signal, and the preliminary detection area is divided using the spectral characteristics, specifically: Acquire the spatiotemporal fusion photoelectric signal of the target area, generate the corresponding spectrum image after time-frequency transformation, and perform frequency band segmentation based on the spectrum image to obtain the main frequency component and energy distribution characteristics; Performing regional interception on the spectrum image according to the energy distribution characteristics to suppress high-frequency noise, obtaining the target core frequency band by intercepting the optimized spectrum image, obtaining the time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band, and converting the time domain distribution into a grayscale spectrum; performing sub-pixel frequency domain positioning in the grayscale spectrum to obtain frequency band edge information, iteratively optimizing the frequency band through least squares ellipse fitting based on the edge information, obtaining the energy center of the optimized ellipse, determining the frequency band width based on the energy center, and generating frequency domain features; Generate spectral features based on the main frequency component, energy distribution characteristics and frequency domain characteristics, use historical spatiotemporal fusion photoelectric signals to extract corresponding spectral features to construct a training data set, and train a convolutional neural network with the training data set to segment the measurement target; Each spectral feature is generated based on sliding window regression, and the corresponding feature area is labeled in the frequency domain to achieve preliminary division of the detection area.

[0016] Preferably, the photoelectric signal of the detection area is intercepted and pattern recognition is performed to complete fine target segmentation, and the spectral characteristics are optimized, specifically: Acquire a preliminary detection area, perform clustering in different frequency domain intervals of the preliminary detection area according to energy density thresholds, and select interval peak points in each frequency domain interval; The energy density threshold is set by extracting the distance between the target optimization ellipse energy centers through the spectral characteristics corresponding to the target, and the distance between the peak point and the nearest frequency domain signal is calculated from high frequency to low frequency to determine whether the distance is less than the energy density threshold; If it is less than, the energy distribution information of the frequency domain interval is updated; if it is greater than, the distance difference is obtained; if the distance difference is less than the preset tolerance, the energy distribution information of the frequency domain interval is updated; The boundary range of the detection area is updated according to the energy distribution information of the frequency domain interval, the clustering of all signals is completed through iteration, the fine target segmentation is completed according to the boundary range of the updated detection area, and the spectral characteristics are optimized.

[0017] Preferably, a feature extraction model is constructed based on a time-frequency analysis method, specifically: A feature extraction model is constructed based on wavelet transform to obtain optimized spectral features, which are then imported into the feature extraction model. The time-frequency features of different frequency domain intervals are extracted using convolution kernels. Hidden nodes of corresponding layers are set according to the time-frequency features of different frequency domain intervals, and fusion is performed through a fully connected layer. A weight distribution layer is set in the feature extraction model, and the attention mechanism is used to obtain the dynamic weights of the time-frequency features in different frequency domain intervals to represent the contribution of the time-frequency features; The weighted time-frequency features are compressed again through convolution and average pooling operations, and the compressed time-frequency features are dimensionality reduced and fused before being imported into the fully connected layer. The probability distribution of the target category is calculated by a normalized exponential function, the target classification result is determined according to the probability distribution, and the target classification result of the target area is combined with the corresponding spectral characteristics to generate the photoelectric measurement data acquisition result of the target area.

[0018] Compared with the prior art, the present invention has the following beneficial effects: During the data acquisition phase, the system acquires data from the photoelectric sensor array and the mobile measurement device, combined with an adaptive spatiotemporal registration strategy, enabling adaptation to complex and changing measurement environments. The spatial grid utilizes an adaptive partitioning method based on the characteristics of the target area, and the time series dynamically adjusts to changes in the measurement environment. This ensures precise alignment of multi-source data under varying terrain, lighting, and other conditions, significantly improving the spatiotemporal consistency and accuracy of the data. Compared to traditional fixed layouts and single acquisition modes, the data acquisition method of this invention enables more comprehensive and accurate acquisition of photoelectric information from the target area, providing a high-quality data foundation for subsequent processing.

[0019] During the data processing process, the present invention adopts a series of innovative methods for spectral feature extraction and target segmentation. Spectral features are obtained through operations such as time-frequency transformation, frequency band segmentation, and ellipse fitting, and are optimized in combination with deep learning technology. This can effectively suppress noise and interference and accurately extract the key spectral features of the measured object. In the fine target segmentation link, based on the analysis of energy density thresholds and frequency domain intervals, through multiple iterative optimizations, high-precision segmentation of targets of different types and characteristics is achieved, avoiding the problems of over-segmentation or under-segmentation, significantly improving the accuracy and reliability of target recognition, and significantly improving the recognition ability of small targets and targets with similar spectra compared to traditional methods.

[0020] In terms of feature extraction and target classification, a feature extraction model built based on time-frequency analysis methods, combined with technologies such as wavelet transforms and attention mechanisms, can deeply explore time-frequency characteristics and dynamically adjust model parameters based on the characteristics of different frequency domain intervals to achieve accurate target classification. This approach fully considers the time-frequency characteristics of photoelectric signals. Compared with traditional fixed-parameter classification models, it has stronger processing capabilities for complex signals and more accurate classification results, and can be effectively applied to a variety of complex measurement scenarios.

[0021] Furthermore, the present invention introduces a dynamic calibration mechanism. By analyzing the energy distribution changes corresponding to the fine target segmentation results in the multi-period photoelectric measurement data of the target area, dynamic calibration parameters are generated in a timely manner. This mechanism monitors the impact of environmental factors and equipment performance changes on measurement data in real time and automatically performs calibration, effectively preventing the accumulation of measurement errors and ensuring the accuracy and stability of long-term measurements. This makes photoelectric measurement technology more practical and adaptable in applications requiring extremely high precision, such as long-term monitoring and industrial automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a working principle diagram of the data acquisition and processing method for photoelectric measurement according to the present invention; Figure 2 Flowchart for spatiotemporal registration and preprocessing of multi-source data; Figure 3 Flowchart for spectral feature acquisition and preliminary detection area division; Figure 4 Flowchart for building a feature extraction model based on time-frequency analysis. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] See also Figures 1-4 The present invention relates to a data acquisition and processing method for photoelectric measurement, and the specific implementation steps are as follows: Obtain photoelectric data of the target area, which includes photoelectric sensor array data and mobile measurement device data. Then, pre-process the acquired photoelectric data and realize spatiotemporal registration of multi-source data. Align the photoelectric data through spatial grids and time series, calculate the spatiotemporal correlation of the grids or sequences corresponding to the photoelectric data, and fuse the grids or sequences whose correlations meet the preset conditions; obtain the phase error in the photoelectric data fusion process, judge the fusion accuracy of the grid or sequence based on the phase error, and perform the next grid fusion when the phase error is within the preset tolerance range until all grids or sequences are fused; perform baseline correction on the fused photoelectric signal using multi-scale decomposition, set dynamic thresholds according to the signal amplitude fluctuations for noise suppression, and obtain the corrected photoelectric signal. The background interference components are removed and normalized; the preprocessed photoelectric signal is obtained, and the local frequency domain features are extracted through an adaptive filtering network. The local frequency domain features are compared for coherence, and the number of feature points is preset. According to the number of feature points, the point where the local frequency domain feature with the highest coherence is obtained as the reference point; the preprocessed photoelectric data is used as the reference signal and the target signal, and the reference point features in the reference signal and the target signal are mapped to obtain the spatiotemporal correlation features, and the spatiotemporal correlation features are matched with the local frequency domain features; the matching weight matrix is ​​obtained, and the signal of the photoelectric data is aligned according to the weight matrix, so as to extract the spatiotemporal fusion photoelectric signal of the target area.

[0025] The spectral characteristics of the measurement object are obtained based on the above-mentioned spatiotemporal fusion photoelectric signal, and the spectral characteristics are used to divide the preliminary detection area. Specifically, the spatiotemporal fusion photoelectric signal of the target area is obtained, and the corresponding spectrum image is generated after time-frequency transformation. The spectrum image is then segmented to obtain the main frequency component and energy distribution characteristics. The spectrum image is regionally intercepted based on the energy distribution characteristics to suppress high-frequency noise. The target core frequency band is obtained by intercepting the optimized spectrum image, and the time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band is obtained. The time domain distribution is converted into a grayscale spectrum. Sub-pixel frequency domain positioning is performed in the grayscale spectrum to obtain band edge information. Based on the edge information, the frequency band is iteratively optimized through least squares ellipse fitting to obtain the energy center of the optimized ellipse. The bandwidth is determined based on the energy center to generate frequency domain characteristics. Spectral characteristics are generated based on the main frequency component, energy distribution characteristics, and frequency domain characteristics. The corresponding spectral characteristics are extracted from the historical spatiotemporal fusion photoelectric signal to construct a training data set. The convolutional neural network is trained with the training data set to segment the measurement target. Each spectral feature is generated by sliding window regression, and the corresponding feature area is labeled in the frequency domain to achieve preliminary division of the detection area.

[0026] The photoelectric signal of the detection area is intercepted for pattern recognition to complete fine target segmentation and optimize the spectral characteristics. The specific operation is to obtain a preliminary detection area, cluster the different frequency domain intervals of the preliminary detection area according to the energy density threshold, and select the interval peak point in each frequency domain interval; extract the target optimization ellipse energy center distance through the spectral characteristics corresponding to the target to set the energy density threshold, use the peak point from high frequency to low frequency to calculate the distance to the nearest frequency domain signal, and judge whether the distance is less than the energy density threshold; if less than, update the energy distribution information of the frequency domain interval; if greater than, obtain the distance difference; if the distance difference is less than the preset tolerance, update the energy distribution information of the frequency domain interval; update the boundary range of the detection area according to the energy distribution information of the frequency domain interval, complete the clustering of all signals through iteration, complete the fine target segmentation according to the boundary range of the updated detection area, and optimize the spectral characteristics.

[0027] A feature extraction model is constructed based on the time-frequency analysis method. The optimized spectral features are used as model input. Target classification is performed through multi-scale decomposition. The photoelectric measurement data acquisition results of the target area are generated based on the classification results and the corresponding spectral features. The specific construction process is to construct a feature extraction model based on wavelet transform, obtain the optimized spectral features, import the feature extraction model, use convolution kernels to extract time-frequency features of different frequency domain intervals, and set the corresponding number of hidden nodes according to the time-frequency features of different frequency domain intervals, and fuse them through a fully connected layer; set a weight distribution layer in the feature extraction model, and use the attention mechanism to obtain the dynamic weights of the time-frequency features of different frequency domain intervals to characterize the contribution of the time-frequency features; the weighted time-frequency features are again compressed through convolution and average pooling operations, and the compressed time-frequency features are dimensionality reduced and fused before being imported into a fully connected layer; the probability distribution of the target category is calculated through a normalized exponential function, and the target classification result is determined based on the probability distribution. The target classification result of the target area is combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results of the target area.

[0028] Finally, the energy distribution changes corresponding to the fine target segmentation results are obtained through multi-period photoelectric measurement data of the target area, and dynamic calibration parameters are generated based on the energy distribution changes. The specific steps are to extract the optimized main frequency component and energy distribution characteristics based on the optimized spectral characteristics, use the optimized main frequency component and energy distribution characteristics to obtain calibration reference points, and generate a calibration reference point set; obtain the fine target segmentation results through multi-period photoelectric measurement data of the target area, and perform data interpolation on the calibration reference point set based on the fine target segmentation results; obtain the data change trend of the calibration reference point set to extract the energy distribution changes, compare the energy distribution changes with the preset calibration range, and generate dynamic calibration parameters when the energy distribution changes exceed the preset calibration range. Embodiment 1:

[0029] During the data acquisition phase, the photoelectric sensor array collects target area data with a fixed layout and sampling frequency, and the mobile measuring device synchronously records information such as position and time, thereby obtaining the target area photoelectric data containing the photoelectric sensor array data and the mobile measuring device data.

[0030] During the spatiotemporal registration phase of multi-source data, the target area is divided into a regular square grid. The side length of each square grid is determined based on the actual size of the target area and the required accuracy. For example, if the target area is 100 square meters and the required measurement accuracy is 1 square meter, the target area is divided into 10×10 square grids. The time series is divided at a fixed sampling interval, which is set based on the frequency characteristics of the photoelectric data to ensure that data changes are effectively captured. When aligning the photoelectric data, the photoelectric data is accurately mapped to the corresponding grid and time point based on the grid coordinate information and the time series timestamps. To calculate spatiotemporal correlation, a correlation coefficient algorithm is used. For each pair of grids or series, the correlation coefficient is calculated to determine whether the correlation meets the preset conditions. The preset conditions can be set based on actual measurement requirements and data characteristics. For example, a correlation coefficient greater than 0.8 is considered to be satisfactory. During the fusion process, phase error is calculated based on the phase information of the data. Phase error is obtained by comparing the phase differences between different data sets. When the phase error is within the preset tolerance, the next grid is fused until all grids or series are fused. The preset tolerance is also determined according to the measurement accuracy requirements.

[0031] For post-fusion photoelectric signal processing, wavelet decomposition is used for multi-scale decomposition during baseline correction. Appropriate wavelet basis functions and decomposition levels are selected to decompose the signal and remove baseline drift. Dynamic thresholds are set based on signal amplitude fluctuations for noise suppression. Statistical methods, such as calculating the mean and standard deviation of the signal amplitude, are used to determine the appropriate threshold range. Portions above the threshold are treated as noise and removed, thereby obtaining the corrected photoelectric signal. Furthermore, a specific algorithm is used to remove background interference components and normalize the signal to maintain a suitable numerical range.

[0032] After obtaining the preprocessed photoelectric signal, the local frequency domain features are extracted through an adaptive filtering network. The adaptive filtering network adopts an adaptive least mean square algorithm, which automatically adjusts the parameters of the filter according to the characteristics of the input signal to adapt to different signal environments. The extracted local frequency domain features are compared for coherence, and the number of feature points is preset, for example, 10 feature points. By calculating the coherence of each local frequency domain feature with these 10 feature points, the point where the local frequency domain feature with the highest coherence is located is obtained as the reference point. Then, the preprocessed photoelectric data is used as the reference signal and the target signal, and the reference point features in the reference signal and the target signal are mapped to obtain the spatiotemporal correlation features, and the spatiotemporal correlation features are matched with the local frequency domain features. The matching weight matrix is ​​obtained through a specific matching algorithm, and the signal alignment of the photoelectric data is performed according to the weight matrix, so as to extract an accurate spatiotemporal fusion photoelectric signal.

[0033] To obtain spectral features and demarcate the detection area, the spatiotemporal fused photoelectric signal undergoes time-frequency transformation. Using a short-time Fourier transform (SFT), the signal is converted from the time domain to the frequency domain, generating a corresponding spectrum image. Frequency bands are segmented based on the frequency distribution characteristics of the spectrum image to obtain the dominant frequency component and energy distribution characteristics. Based on the energy distribution characteristics, the spectrum image is regionally clipped to suppress high-frequency noise. The spectrum image is optimized by adjusting the clipping range multiple times to obtain the target core frequency band. The time domain distribution of the spatiotemporal fused photoelectric signal corresponding to the core frequency band is obtained and converted into a grayscale spectrum. Sub-pixel frequency domain positioning is performed within the grayscale spectrum, using a specific positioning algorithm to obtain band edge information. Based on this edge information, the frequency band is iteratively optimized using a least-squares ellipse fitting algorithm. The ellipse parameters are adjusted with each iteration to obtain the energy center of the optimized ellipse. The frequency band width is determined based on the energy center, generating frequency domain features. Spectral features are generated based on the dominant frequency component, energy distribution characteristics, and frequency domain features. A large number of historical spatiotemporal fused photoelectric signals and their corresponding spectral features are collected to construct a training dataset, ensuring that the training dataset covers a wide range of measurement conditions and target characteristics, ensuring diversity and representativeness. The convolutional neural network is trained on the training data set to segment the measurement target. Spectral features are generated based on sliding window regression, and the corresponding feature areas are annotated in the frequency domain to achieve preliminary division of the detection area.

[0034] During the fine target segmentation and spectral feature optimization phase, after obtaining the preliminary detection area, energy density thresholds are appropriately set in different frequency domain intervals within the preliminary detection area based on the target's spectral characteristics and actual measurement results. Clustering is then performed, and interval peak points are selected within each frequency domain interval. The distance between the energy centers of the target optimization ellipse is extracted from the target's corresponding spectral characteristics, and this is used to set the energy density threshold. The distance from the peak point to the nearest neighboring frequency domain signal is calculated from high to low frequency, and a determination is made as to whether this distance is less than the energy density threshold. If so, the energy distribution information for the frequency domain interval is updated. If so, the distance difference is obtained. If the distance difference is less than a preset tolerance, the energy distribution information for the frequency domain interval is also updated. The preset tolerance is set based on the measurement accuracy requirements. The detection area boundaries are updated based on the energy distribution information for the frequency domain intervals. Clustering of all signals is completed through multiple iterations. Fine target segmentation is then completed based on the updated detection area boundaries, and the spectral features are optimized.

[0035] When constructing the feature extraction model, a wavelet transform is used. Based on the time-frequency features of different frequency domain intervals, the size and number of convolution kernels are appropriately set to effectively extract time-frequency features. Based on the extracted time-frequency features, the corresponding number of hidden nodes is accurately set, and fusion is performed through a fully connected layer. A weight distribution layer is set in the feature extraction model, and the attention mechanism is used to obtain the dynamic weights of the time-frequency features of different frequency domain intervals. These weights can accurately represent the contribution of the time-frequency features. The weighted time-frequency features are further compressed through convolution and average pooling operations to reduce the feature dimension. The compressed time-frequency features are then dimensionality-reduced and fused and imported into the fully connected layer. The probability distribution of the target category is calculated using a normalized exponential function. The target classification result is determined based on the probability distribution. The target classification result of the target area is combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results of the target area.

[0036] When generating dynamic calibration parameters, the optimized dominant frequency component and energy distribution characteristics are extracted based on the optimized spectral features. These characteristics are used to obtain calibration reference points and generate a calibration reference point set. Fine target segmentation results are obtained by multi-period photoelectric measurement data of the target area. Based on this fine target segmentation, data interpolation is performed on the calibration reference point set to supplement missing data points. The data trend of the calibration reference point set is carefully analyzed to extract energy distribution changes. These energy distribution changes are compared with the preset calibration range. When the energy distribution changes exceed the preset calibration range, dynamic calibration parameters are generated. Example 2:

[0037] The data acquisition process is consistent with the overall solution, and the photoelectric data of the target area is obtained through the photoelectric sensor array and the mobile measuring device.

[0038] During the spatiotemporal registration of multi-source data, the target area is divided into irregular triangular grids. This division method is determined based on factors such as the target area's topography and the distribution of key measurement areas. For example, for target areas with complex terrain and numerous irregular boundaries, triangular grids can be flexibly divided according to the actual terrain contours, ensuring that each triangular grid better fits the actual measurement area. The time series is unevenly divided based on the requirements of the measurement task. During periods of high or critical photoelectric data fluctuations, smaller time intervals are used; during periods of moderate data fluctuations, larger time intervals are used. This ensures data validity while reducing data processing. Coordinate transformation and temporal interpolation are used to align photoelectric data. For spatial coordinates, data collected by different sensors are unified into a common coordinate system using a coordinate transformation algorithm. For the temporal dimension, a temporal interpolation algorithm is used to interpolate data with different sampling frequencies to a common time node, accurately mapping the data to the corresponding grid and time point. To calculate spatiotemporal correlation, a mutual information algorithm is used to determine correlation by calculating the mutual information value between different data sets. The mutual information value reflects the degree of dependence between two data sequences. An appropriate mutual information threshold is set. When the mutual information value exceeds the threshold, the data are considered to meet the required correlation. During the fusion process, the calculation of phase error takes into account the frequency and phase variation trends of the data. By analyzing the phase differences at different frequencies and the phase variation over time, the phase error is accurately calculated. For baseline correction, multi-scale decomposition uses empirical mode decomposition, which decomposes the signal into multiple intrinsic mode functions based on its inherent characteristics. The dynamic threshold is set based on the local characteristics of the signal. By analyzing the signal characteristics within a local time period or spatial region, such as the amplitude variation range and frequency distribution, an appropriate dynamic threshold is determined to effectively suppress noise. To extract local frequency domain features, the adaptive filtering network uses a recursive least squares algorithm. This algorithm continuously updates the filter parameters, enabling the filter to quickly adapt to signal changes, thereby accurately extracting local frequency domain features. During signal registration, the calculation of the matching weight matrix is ​​optimized, for example by introducing a regularization term, to avoid overfitting during the matrix calculation process and improve signal registration accuracy.

[0039] During the spectral feature acquisition and detection area division phase, the spatiotemporal fused photoelectric signal undergoes a time-to-frequency transformation using the Gabor transform. The Gabor transform analyzes signals simultaneously in the time and frequency domains, generating a spectrum image with excellent time-frequency resolution. Frequency band segmentation is performed based on the energy distribution of the spectrum image. By analyzing the energy concentration and distribution within the spectrum image, the frequency band boundaries are determined, thereby obtaining the dominant frequency component and energy distribution characteristics. Based on the energy distribution characteristics, the spectrum image is regionally cropped, focusing on the energy-concentrated areas and suppressing high-frequency noise. The spectrum image is optimized by adjusting the cropping range and parameters multiple times to obtain the target core frequency band. The time domain distribution of the spatiotemporal fused photoelectric signal corresponding to the core frequency band is obtained and converted into a grayscale spectrum. Sub-pixel frequency domain positioning is performed within the grayscale spectrum using a specific positioning algorithm, such as a gradient-based positioning algorithm, to obtain band edge information. Based on this edge information, the frequency band is iteratively optimized using a least-squares ellipse fitting algorithm. The ellipse parameters are adjusted based on the fitting error at each iteration to obtain the energy center of the optimized ellipse. The band width is determined based on the energy center, generating frequency domain features. Spectral features are generated based on the dominant frequency components, energy distribution characteristics, and frequency domain features. When constructing the training dataset, historical data is screened and preprocessed to remove abnormal data, such as erroneous data caused by sensor failure or external interference, to ensure dataset quality. The training dataset is constructed by extracting the corresponding spectral features from the filtered historical spatiotemporal fusion photoelectric signals. A convolutional neural network is trained on the training dataset to segment the measurement targets. Spectral features are generated using sliding window regression, and corresponding feature regions are labeled in the frequency domain to achieve preliminary demarcation of the detection area.

[0040] During the fine target segmentation and spectral feature optimization phase, after obtaining a preliminary detection area, the energy density threshold is adjusted based on the target's shape and size in different frequency domain intervals within the preliminary detection area. Clustering is then performed, and interval peak points are selected within each frequency domain interval. The distance between the energy centers of the target optimization ellipse is extracted using the target's corresponding spectral features, and this is used to set the energy density threshold. The distance from the peak point to the nearest neighboring frequency domain signal is calculated from high to low frequency, and a determination is made as to whether this distance is less than the energy density threshold. If so, the energy distribution information for the frequency domain interval is updated. If so, the distance difference is obtained. If the distance difference is less than a preset tolerance, the energy distribution information for the frequency domain interval is also updated. The preset tolerance is set based on the measurement accuracy requirements. The detection area boundaries are updated based on the energy distribution information for the frequency domain interval. Through multiple iterations and optimization, the detection area boundaries are gradually refined, and all signals are clustered. Fine target segmentation is then completed based on the updated detection area boundaries, and the spectral features are optimized.

[0041] When constructing the feature extraction model, improvements are made based on the wavelet transform. The shape and parameters of the convolution kernel are adjusted, for example, using kernels of different sizes and shapes, to accommodate the time-frequency feature extraction requirements of different frequency domain intervals. The hidden node connections are optimized, and a more complex connection structure is adopted to enhance the model's ability to express time-frequency features. An attention mechanism is used to better highlight important time-frequency features. By calculating the weights of different time-frequency features, the model focuses more on important features during training and prediction. The weighted time-frequency features are further compressed through convolution and average pooling operations. The compressed time-frequency features are then dimensionality reduced and fused and imported into the fully connected layer. The probability distribution of the target category is calculated using a normalized exponential function. The target classification result is determined based on the probability distribution. The target classification result of the target area is combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results of the target area.

[0042] When generating dynamic calibration parameters, the optimized dominant frequency component and energy distribution characteristics are extracted based on the optimized spectral features. These characteristics are used to obtain calibration reference points and generate a calibration reference point set. Fine target segmentation results are obtained by multi-period photoelectric measurement data of the target area. Based on this fine target segmentation result, data interpolation is performed on the calibration reference point set to supplement missing data points. The calibration reference point set is analyzed by comprehensively considering multiple factors, such as measurement time and environmental conditions, to more accurately extract energy distribution changes. This energy distribution change is then compared with the preset calibration range. Dynamic calibration parameters are generated when the energy distribution change exceeds the preset calibration range. Example 3:

[0043] Starting from data collection, the photoelectric data of the target area is still obtained with the help of photoelectric sensor arrays and mobile measuring devices. The layout of the sensor array and the moving path of the measuring device will be pre-planned and set according to the characteristics of the target area.

[0044] During the spatiotemporal registration of multi-source data, a hexagonal grid is used to divide the target area. Compared to conventional grids, hexagonal grids offer better spatial balance and symmetry, making them particularly suitable for measurement scenarios requiring high isotropy. The side length of the hexagonal grid is determined based on the target area's scope and the required measurement accuracy. For example, if the target area is a circle, to ensure uniform measurement across all directions, it can be divided into tightly packed hexagonal grids. The side length of each hexagonal grid is determined based on the required measurement resolution. For example, a hexagonal grid with a side length of 1 meter can achieve a measurement accuracy of approximately 1 square meter. Time series are divided based on event triggering. New time series nodes are triggered when significant changes in photoelectric data occur, when the measurement device enters a specific area, or when preset measurement conditions are met. When aligning photoelectric data, spatial geometric relationships are leveraged to accurately map each data point to its corresponding grid cell by calculating its coordinate position within the hexagonal grid. Furthermore, data is associated with corresponding time nodes based on the logical relationship between temporal events. When calculating spatiotemporal correlation, a dynamic time warping algorithm is used. This algorithm can effectively calculate the degree of similarity between different data series even when the data time series exhibits nonlinear variations. During the calculation process, a distance matrix is ​​constructed to find the optimal time warping path, thereby determining the correlation between the data. During the fusion process, phase error is calculated based on the data's phase change rate. Phase error values ​​are calculated by analyzing the phase change trends of data at adjacent time points or adjacent grids. Baseline correction is performed using a combination of Fourier transform and wavelet transform. First, the Fourier transform is used to convert the signal from the time domain to the frequency domain, analyzing and preliminarily processing the overall frequency components of the signal. Then, the wavelet transform is used to perform multi-scale decomposition of the signal, further refining its features and removing baseline drift and high-frequency noise. A dynamic threshold is dynamically adjusted based on the signal's changing trend. By monitoring the amplitude fluctuation range of the signal in real time and combining statistical methods such as moving average and standard deviation, an appropriate threshold is determined to achieve effective noise suppression. To extract local frequency domain features, an adaptive filtering network based on a neural network is constructed using an adaptive algorithm. The network learns the characteristic patterns of signals through training and automatically adjusts the filter parameters to adapt to the needs of local frequency domain feature extraction in different signal environments. During signal registration, by introducing constraints such as spatial position constraints and temporal sequence constraints, the calculation process of the matching weight matrix is ​​optimized to ensure the precise registration of optical and electrical data, thereby extracting accurate spatiotemporal fusion optical and electrical signals.

[0045] To obtain spectral features and demarcate detection areas, the time-frequency transformation of the spatiotemporal fused photoelectric signal is performed using a fractional Fourier transform. As a generalized Fourier transform, the fractional Fourier transform can analyze signals at different fractional orders. It is suitable for processing signals with nonlinear frequency modulation characteristics and more effectively captures the signal's time-frequency characteristics than the traditional Fourier transform. Frequency band segmentation is performed based on the frequency characteristics of the spectrum image and the characteristics of the target. By analyzing the distribution of different frequency components in the spectrum image and combining the target's inherent frequency domain characteristics, such as energy concentration areas at specific frequencies, the frequency band boundaries are determined to obtain the dominant frequency component and energy distribution characteristics. Based on the energy distribution characteristics, the spectrum image is regionally cropped, focusing on retaining frequency regions relevant to the target and suppressing high-frequency noise. By repeatedly adjusting the cropping parameters, the spectrum image is optimized to obtain the target's core frequency band. The time domain distribution of the spatiotemporal fused photoelectric signal corresponding to the core frequency band is obtained and converted into a grayscale spectrum. Sub-pixel frequency domain positioning is performed within the grayscale spectrum. High-precision positioning algorithms, such as those based on image edge detection and morphological processing, are employed to obtain band edge information. Based on this edge information, the band is iteratively optimized using least-squares ellipse fitting. The ellipse parameters are adjusted based on the fitting error at each iteration until a satisfactory fit is achieved. The energy center of the optimized ellipse is then determined. The band width is determined based on the energy center, and frequency domain features are generated. Spectral features are generated based on the dominant frequency component, energy distribution characteristics, and frequency domain features. When constructing the training dataset, data augmentation techniques are employed to expand the dataset. For example, operations such as translation, scaling, and rotation are performed on the historical spatiotemporal fused photoelectric signals to simulate different measurement angles and positions. Alternatively, appropriate amounts of Gaussian noise can be added to simulate interference conditions encountered in actual measurements, thereby increasing the dataset's diversity and improving the model's generalization capabilities. A convolutional neural network is trained using this expanded training dataset to segment the measurement target. Spectral features are generated using sliding window regression, and corresponding feature regions are labeled in the frequency domain to achieve preliminary delineation of the detection area.

[0046] During fine target segmentation and spectral feature optimization, after obtaining a preliminary detection area, energy density thresholds are set based on the target's material and optical properties within different frequency domain intervals within the preliminary detection area. Clustering is then performed, and peak points are selected within each frequency domain interval. Targets of different materials exhibit unique energy distribution characteristics in the spectrum. By analyzing the spectral characteristics corresponding to the target material, an appropriate energy density threshold is set to accurately identify the target. The distance between the energy centers of the target optimization ellipse is extracted from the target's corresponding spectral characteristics, and this is used to set the energy density threshold. The distance from the peak point to the nearest neighboring frequency domain signal is calculated from high to low frequencies to determine whether the distance is less than the energy density threshold. If so, the energy distribution information for the frequency domain interval is updated. If so, the distance difference is calculated. If the distance difference is less than a preset tolerance, the energy distribution information for the frequency domain interval is also updated. The preset tolerance is set based on the measurement accuracy requirements. The detection area boundaries are updated based on the energy distribution information within the frequency domain interval. Through multiple complex clustering and iterative algorithms, the detection area boundaries are gradually refined, completing clustering of all signals. Fine target segmentation is then performed based on the updated detection area boundaries, and the spectral features are optimized.

[0047] When constructing the feature extraction model, a deep neural network model is built based on wavelet transforms. The number of network layers and complexity are increased, for example, by building a multi-layer neural network structure consisting of multiple convolutional layers, pooling layers, and fully connected layers to enhance the model's ability to extract and represent time-frequency features. Convolution kernels are used to extract time-frequency features in different frequency domain intervals. Hidden nodes are assigned to the corresponding layers based on the time-frequency features in these frequency domain intervals, and these features are fused through fully connected layers. A weight assignment layer is implemented within the feature extraction model, utilizing an attention mechanism to more accurately weight the time-frequency features. By calculating the importance of different time-frequency features in target classification and assigning them corresponding weights, the model focuses more on key features during training and prediction. The weighted time-frequency features are further compressed through convolution and average pooling operations. The compressed time-frequency features are then fused and fed into a fully connected layer. The probability distribution of target categories is calculated using a normalized exponential function. The target classification result is determined based on this probability distribution. The target classification result is combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results for the target area.

[0048] When generating dynamic calibration parameters, the optimized dominant frequency components and energy distribution characteristics are extracted based on the optimized spectral features. These characteristics are used to obtain calibration reference points and generate a calibration reference point set. Fine target segmentation results are obtained by multi-period photoelectric measurement data of the target area. Based on the fine target segmentation results, data interpolation is performed on the calibration reference point set to supplement missing data points. Machine learning algorithms such as regression analysis and decision trees are used to analyze the calibration reference point set, exploring potential relationships and change patterns between the data to more accurately predict energy distribution changes. The energy distribution change is compared with the preset calibration range. When the energy distribution change exceeds the preset calibration range, dynamic calibration parameters are generated. Embodiment 4:

[0049] During the data acquisition phase, the photoelectric sensor array is arranged around or inside the target area according to a specific distribution pattern, and the mobile measurement device performs mobile measurement of the target area according to a pre-planned path, thereby obtaining photoelectric data of the target area including photoelectric sensor array data and mobile measurement device data.

[0050] When entering the spatiotemporal registration phase of multi-source data, the spatial grid adopts a hybrid grid structure, including square, triangular and hexagonal grids. When dividing, the appropriate grid type is selected according to the shape, complexity and measurement requirements of different parts of the target area. For example, a square grid is used for areas with regular shapes and uniform measurement accuracy requirements; a triangular grid is used for areas with irregular boundaries to better fit the boundaries; and a hexagonal grid is used for areas that require isotropic measurements. The time series adopts a hybrid division method, combining fixed intervals and event triggering. Fixed time interval sampling is used in stages where data changes are relatively stable, and event-triggered sampling is used in periods where data may mutate or are of particular concern, such as when the sensor detects a mutation in an environmental parameter, triggering sampling.

[0051] When aligning optoelectronic data, a combination of mapping methods is used. For spatial coordinates, corresponding coordinate transformation relationships are established for different types of grids, and sensor data is mapped to a unified coordinate system. For the time dimension, fixed-interval sampling data and event-triggered sampling data are unified using a time interpolation algorithm. When calculating spatiotemporal correlation, a combined algorithm is used, combining the correlation coefficient algorithm, the mutual information algorithm, and the dynamic time warping algorithm. First, the correlation coefficient algorithm is used to preliminarily screen out data pairs with high correlation. The mutual information algorithm is then used to further measure the degree of dependence between the data. For data with large time series differences, the dynamic time warping algorithm is used to calculate their similarity.

[0052] In the fusion process, the calculation of phase error comprehensively considers multiple dimensions of data information. For the The spatial grid in The phase value of the data at each time point, the phase error The calculation formula is:

[0053] in is the total number of spatial grids, is the total number of time points, For the The average phase value of all spatial grid data at each time point. For baseline correction, multiscale decomposition uses a multiscale morphological decomposition method, performing opening and closing operations on the signal using structuring elements of different scales to remove baseline drift. The dynamic threshold is set based on the local and global characteristics of the signal. The mean and variance of the local signal region are first calculated, and then the threshold is determined based on the statistical characteristics of the global signal.

[0054] When extracting local frequency domain features, the adaptive filtering network uses a hybrid adaptive algorithm that combines the strengths of the least mean square algorithm and the recursive least squares algorithm, automatically switching algorithms based on the signal's stationarity and rate of change. During signal registration, the accuracy of the matching weight matrix is ​​improved by optimizing the algorithm and adding constraints such as spatial position and temporal order. For example, data from spatially adjacent grids is assigned a higher weight during registration, and temporally continuous data is required to maintain sequential consistency.

[0055] When acquiring spectral features and dividing the detection area, the spatiotemporal fusion photoelectric signal is subjected to time-frequency transformation, using a combination of various transformation methods. For example, a short-time Fourier transform is first performed to obtain the signal's preliminary time-frequency distribution, followed by a fractional-order Fourier transform to refine the analysis of specific frequency components. Frequency band segmentation is performed based on the spectral features corresponding to various characteristics of the target, such as the target's material, shape, and motion state. The spectrum image is regionally intercepted based on the energy distribution characteristics to suppress high-frequency noise and obtain the target's core frequency band. The time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band is then converted into a grayscale spectrum. Sub-pixel frequency domain positioning is performed in the grayscale spectrum to obtain band edge information. The frequency band is iteratively optimized through least-squares ellipse fitting to obtain the energy center of the optimized ellipse. The bandwidth is determined based on the energy center to generate frequency domain features.

[0056] When constructing the training dataset, feature selection and dimensionality reduction are performed on the data. Methods such as principal component analysis (PCA) are used to remove redundant features, retain the most valuable features for target classification, reduce data dimensionality, and improve training efficiency. The training dataset is constructed by extracting corresponding spectral features from the processed historical spatiotemporal fusion photoelectric signals. A convolutional neural network is trained on this training dataset to segment the measurement targets. Spectral features are generated using sliding window regression, and corresponding feature regions are labeled in the frequency domain to achieve preliminary division of the detection area.

[0057] During the fine target segmentation and spectral feature optimization phase, after obtaining the preliminary detection area, the energy density threshold is dynamically adjusted within the different frequency domain intervals of the preliminary detection area based on the target's various attributes (such as material, size, and reflectivity). Clustering operations are performed and interval peak points are selected. The energy density threshold is set based on the distance between the energy centers of the target optimization ellipse extracted from the target's corresponding spectral features. The distance to the nearest frequency domain signal is calculated from high to low frequency using the peak point to determine whether the distance is less than the energy density threshold. If so, the energy distribution information of the frequency domain interval is updated; if so, the distance difference is obtained. If the distance difference is less than the preset tolerance, the energy distribution information of the frequency domain interval is updated. The boundary range of the detection area is updated based on the energy distribution information of the frequency domain interval. Through complex iterative and optimization strategies, all signals are clustered. Fine target segmentation is completed based on the boundary range of the updated detection area, and the spectral features are optimized.

[0058] When constructing the feature extraction model, a heterogeneous neural network model is built based on wavelet transforms, integrating the advantages of convolutional neural networks (CNNs) and recurrent neural networks (RNNs). CNNs are used to extract spatial information from time-frequency features, while RNNs are used to process the temporal series information of these features. Convolution kernels are used to extract time-frequency features in different frequency domain intervals. Hidden nodes are assigned to the corresponding number of layers based on the features, and fusion is performed through fully connected layers. A weight distribution layer is incorporated into the feature extraction model, utilizing an attention mechanism to more comprehensively capture the importance of time-frequency features. The weighted time-frequency features are then subjected to convolution, pooling, dimensionality reduction, and fusion. The probability distribution of the target category is calculated using a normalized exponential function. The target classification result is then determined based on the probability distribution. The target classification result is then combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results for the target area.

[0059] When generating dynamic calibration parameters, the optimized dominant frequency component and energy distribution characteristics are extracted based on the optimized spectral features. These characteristics are used to obtain calibration reference points and generate a calibration reference point set. Fine target segmentation results are obtained by analyzing the target area's multi-period photoelectric measurement data. Data interpolation is performed on the calibration reference point set based on these fine target segmentation results. Using big data analysis techniques such as association rule mining and temporal pattern analysis, the calibration reference point set is deeply analyzed to more accurately determine energy distribution changes. These energy distribution changes are then compared with the preset calibration range. Dynamic calibration parameters are generated when the energy distribution changes exceed the preset calibration range. Example 5:

[0060] Data collection is carried out based on the actual characteristics of the target area. The deployment of the photoelectric sensor array is not a fixed pattern. Instead, it is targeted through preliminary surveys of the target area, combined with the target's likely location, activity patterns, and environmental factors. For example, in mountainous areas with complex terrain, to fully capture the target's photoelectric information, the sensor array will be dispersed along key locations such as valleys and ridges. The mobile measurement device will also plan flexible movement paths based on the preset measurement tasks and the target's activity range, ensuring that photoelectric data from different angles and positions can be obtained during movement. Ultimately, the photoelectric sensor array data and the mobile measurement device data are fully acquired and used as raw data for subsequent processing.

[0061] When it comes to spatiotemporal registration of multi-source data, the spatial grid utilizes an adaptive meshing method based on the characteristics of the target area. This method utilizes advanced image recognition and analysis technologies to process the image data of the target area. First, it identifies the various topographic features and object distribution within the target area. Then, it dynamically generates a grid based on the complexity of these features and the required measurement accuracy. In areas with dense objects and diverse features, the grid is finer to ensure accurate capture of changes in the photoelectric data. In relatively flat areas with simpler features, the grid is more sparse, thereby ensuring measurement accuracy while reducing data processing. The time series also adaptively adjusts to changes in the measurement environment. By monitoring environmental parameters such as light intensity, temperature, and wind speed in real time, the system automatically adjusts the time series sampling interval when significant changes in these parameters may affect photoelectric data acquisition. For example, when light intensity fluctuates dramatically, the sampling interval is shortened to ensure timely capture of photoelectric data fluctuations caused by the change.

[0062] When aligning optoelectronic data, an intelligent algorithm is used for dynamic mapping. This algorithm, based on a deep learning model, learns from a large amount of historical data to understand the mapping patterns of optoelectronic data at different spatial locations and time points. For newly acquired data, the model automatically and accurately maps it to the corresponding spatial grid and time point based on the data's characteristics. When calculating spatiotemporal correlations, a correlation calculation method based on deep learning is used. A specialized deep learning network model is constructed, taking different optoelectronic data as input. Through multi-layer neuron calculations and feature extraction within the network, the correlation value between the data is output. During training, the network model continuously optimizes parameters to improve the accuracy of data correlation calculations. During the fusion process, the calculation of phase error is based on the deep learning model's prediction and analysis of the data phase. The model learns the phase variation patterns in historical data, predicts the phase of new data, and then compares the predicted value with the actual value to calculate the phase error.

[0063] For baseline correction, multi-scale decomposition uses an adaptive multi-scale decomposition method. This method uses an algorithm to automatically determine and select the most appropriate decomposition method and parameters based on the signal's characteristics. For example, for signals with significant periodicity, decomposition methods related to Fourier transform are selected; for non-stationary signals, methods such as wavelet transform are used. Dynamic thresholds are optimized using a machine learning algorithm. A machine learning model is established that takes various signal characteristics, such as amplitude, frequency, and trend, as input and outputs a dynamic threshold. During model training, the model continuously adjusts parameters based on the relationship between different signal characteristics and the optimal threshold to achieve precise threshold optimization, effectively suppressing noise. For extracting local frequency domain features, an adaptive filtering network based on deep learning is used. This network, composed of multiple deep learning modules, is trained on large amounts of data to automatically learn the signal's frequency domain characteristics and adjust filtering parameters in real time based on changes in the input signal, accurately extracting local frequency domain features. For signal registration, the deep learning model is used to generate a matching weight matrix. By learning successful cases of signal registration in historical data, the model grasps the relationship between different signal features and weight matrices, thereby generating accurate matching weight matrices for new optoelectronic data, achieving high-precision signal registration, and extracting high-quality spatiotemporal fusion optoelectronic signals.

[0064] To obtain spectral features and demarcate detection areas, a deep learning-based time-frequency transformation method is used to transform the spatiotemporally fused photoelectric signals. A deep learning network is constructed and, after training with a large number of time-frequency transformation samples, it can directly transform the input photoelectric signals and generate accurate spectrum images. Frequency band segmentation is also performed based on the deep learning model's analysis of the spectrum images. By learning the frequency band characteristics of different targets in a large number of spectrum images, the model can automatically identify the target-related frequency bands in the spectrum images and determine their boundaries, thereby obtaining the dominant frequency components and energy distribution characteristics. Based on the energy distribution characteristics, the spectrum images are regionally segmented to suppress high-frequency noise and obtain the target's core frequency band. The time domain distribution of the spatiotemporally fused photoelectric signal corresponding to the core frequency band is obtained and converted into a grayscale spectrum. Sub-pixel frequency domain localization is performed within the grayscale spectrum to obtain band edge information. The deep learning model then analyzes the grayscale spectrum to accurately obtain band edge information by identifying features such as edges and textures within the image. Based on the edge information, the frequency band is iteratively optimized through least squares ellipse fitting to obtain the energy center of the optimized ellipse. The bandwidth is determined based on the energy center to generate frequency domain features. Spectral features are generated based on the main frequency component, energy distribution characteristics, and frequency domain features.

[0065] When constructing the training dataset, the system continuously updates it using online learning. During system operation, each new set of valid data is added to the training dataset. The system then filters and organizes the existing dataset based on the characteristics of the new data, removing redundant or no longer representative data. The convolutional neural network is trained using this updated training dataset, enabling the network to continuously adapt to new data features and improve the accuracy of target segmentation. Spectral features are generated using sliding window regression, and corresponding feature regions are annotated in the frequency domain to achieve preliminary delineation of the detection area.

[0066] During fine target segmentation and spectral feature optimization, after obtaining a preliminary detection area, the energy density thresholds are dynamically set within different frequency domain intervals within the preliminary detection area based on the target analysis results of the deep learning model. The deep learning model comprehensively analyzes the target's spectral characteristics, shape, material, and other aspects to determine the appropriate energy density threshold for each frequency domain interval. Clustering is then performed within each frequency domain interval to select the interval peak point. The distance from the peak point to the nearest neighboring frequency domain signal is calculated from high to low frequency to determine whether the distance is less than the energy density threshold. If so, the energy distribution information for the frequency domain interval is updated. If so, the distance difference is calculated. If the distance difference is less than the preset tolerance, the energy distribution information for the frequency domain interval is updated. The detection area boundaries are updated based on the energy distribution information for the frequency domain interval. Multiple iterations and optimizations are performed using the deep learning algorithm to complete clustering of all signals. Fine target segmentation is then performed based on the updated detection area boundaries, and the spectral features are optimized.

[0067] When constructing the feature extraction model, a complex neural network model is built based on deep learning. The model contains multiple network layers with different functions, such as convolutional layers for extracting time-frequency features, pooling layers for reducing data dimensionality, and recurrent layers for processing time series information. Attention mechanisms and other deep learning techniques, such as self-attention and residual connections, are used to better extract and fuse time-frequency features. During model training, network parameters are continuously adjusted using a backpropagation algorithm, enabling the model to accurately extract features of various targets. The weighted time-frequency features are subjected to a series of operations, such as convolution, pooling, and dimensionality reduction, before being fed into a fully connected layer. The probability distribution of the target category is calculated using a normalized exponential function. The target classification result is then determined based on the probability distribution. The target classification result is then combined with the corresponding spectral features to generate the photoelectric measurement data acquisition results for the target area.

[0068] When generating dynamic calibration parameters, the optimized dominant frequency component and energy distribution characteristics are extracted based on the optimized spectral features. These characteristics are used to obtain calibration reference points and generate a calibration reference point set. Fine target segmentation results are obtained by multi-period photoelectric measurement data of the target area. Data interpolation is performed on the calibration reference point set based on the fine target segmentation results. The calibration reference point set is analyzed using a deep learning prediction model. This model predicts future energy distribution changes by learning from the variation patterns of the calibration reference point set in historical data. The predicted energy distribution change is compared with the preset calibration range. If the energy distribution change exceeds the preset calibration range, dynamic calibration parameters are generated.

[0069] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0070] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data acquisition and processing method for photoelectric measurement, characterized in that: The following steps are involved: Acquire optoelectronic data of the target area, including optoelectronic sensor array data and mobile measurement device data; preprocess the acquired optoelectronic data and implement spatiotemporal registration of multi-source data to extract spatiotemporal fusion optoelectronic signals of the target area; Acquiring spectral characteristics of the measurement object based on the spatiotemporal fusion photoelectric signal, dividing a preliminary detection area using the spectral characteristics, intercepting the photoelectric signal of the detection area to perform pattern recognition to complete fine target segmentation, and optimizing the spectral characteristics; A feature extraction model is constructed based on the time-frequency analysis method. The optimized spectral features are used as the model input. The target is classified through multi-scale decomposition. The photoelectric measurement data acquisition results of the target area are generated based on the classification results and the corresponding spectral features. The energy distribution change corresponding to the fine target segmentation result is obtained through the multi-time period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change.

2. The data acquisition and processing method for photoelectric measurement according to claim 1, characterized in that: The acquired photoelectric data is preprocessed and multi-source data spatiotemporal registration is achieved to extract the spatiotemporal fusion photoelectric signals of the target area. Specifically: Aligning the photoelectric data by spatial grids and time series, obtaining the spatiotemporal correlation of the photoelectric data corresponding to the grids or sequences, and fusing the grids or sequences whose correlations meet preset conditions; Obtaining the phase error during the photoelectric data fusion process, judging the fusion accuracy of the grid or sequence based on the phase error, and performing the next grid fusion when the phase error is within a preset tolerance range until all grids or sequences are fused; The fused photoelectric signal is subjected to baseline correction using multi-scale decomposition, and a dynamic threshold is set according to the signal amplitude fluctuation to suppress noise. The corrected photoelectric signal is obtained, and the background interference component is eliminated and normalized. Obtaining the preprocessed photoelectric signal, extracting local frequency domain features through an adaptive filtering network, performing coherence comparison on the local frequency domain features, presetting a number of feature points, and obtaining the point where the local frequency domain feature with the highest coherence is located as a reference point based on the number of feature points; Using the preprocessed photoelectric data as a reference signal and a target signal, mapping the fiducial point features in the reference signal and the target signal to obtain spatiotemporal correlation features, and matching the spatiotemporal correlation features with local frequency domain features; A matching weight matrix is ​​obtained, and signal registration of the photoelectric data is performed according to the weight matrix, and the spatiotemporal fusion photoelectric signal of the registered target area is extracted.

3. The data acquisition and processing method for photoelectric measurement according to claim 1, characterized in that: The spectral characteristics of the measurement object are obtained according to the spatiotemporal fusion photoelectric signal, and the preliminary detection area is divided using the spectral characteristics, specifically: Acquire the spatiotemporal fusion photoelectric signal of the target area, generate the corresponding spectrum image after time-frequency transformation, and perform frequency band segmentation based on the spectrum image to obtain the main frequency component and energy distribution characteristics; Performing regional interception on the spectrum image according to the energy distribution characteristics to suppress high-frequency noise, obtaining the target core frequency band by intercepting the optimized spectrum image, obtaining the time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band, and converting the time domain distribution into a grayscale spectrum; performing sub-pixel frequency domain positioning in the grayscale spectrum to obtain frequency band edge information, iteratively optimizing the frequency band through least squares ellipse fitting based on the edge information, obtaining the energy center of the optimized ellipse, determining the frequency band width based on the energy center, and generating frequency domain features; Generate spectral features based on the main frequency component, energy distribution characteristics and frequency domain characteristics, use historical spatiotemporal fusion photoelectric signals to extract corresponding spectral features to construct a training data set, and train a convolutional neural network with the training data set to segment the measurement target; Each spectral feature is generated based on sliding window regression, and the corresponding feature area is labeled in the frequency domain to achieve preliminary division of the detection area.

4. The data acquisition and processing method for photoelectric measurement according to claim 1, characterized in that: The photoelectric signal of the detection area is intercepted and pattern recognition is performed to complete fine target segmentation, and the spectral characteristics are optimized, specifically: Acquire a preliminary detection area, perform clustering in different frequency domain intervals of the preliminary detection area according to energy density thresholds, and select interval peak points in each frequency domain interval; The energy density threshold is set by extracting the distance between the target optimization ellipse energy centers through the spectral characteristics corresponding to the target, and the distance between the peak point and the nearest frequency domain signal is calculated from high frequency to low frequency to determine whether the distance is less than the energy density threshold; If it is less than, the energy distribution information of the frequency domain interval is updated; if it is greater than, the distance difference is obtained; if the distance difference is less than the preset tolerance, the energy distribution information of the frequency domain interval is updated; The boundary range of the detection area is updated according to the energy distribution information of the frequency domain interval, the clustering of all signals is completed through iteration, the fine target segmentation is completed according to the boundary range of the updated detection area, and the spectral characteristics are optimized.

5. The data acquisition and processing method for photoelectric measurement according to claim 1, characterized in that: A feature extraction model is constructed based on the time-frequency analysis method, specifically: A feature extraction model is constructed based on wavelet transform to obtain optimized spectral features, which are then imported into the feature extraction model. The time-frequency features of different frequency domain intervals are extracted using convolution kernels. Hidden nodes of corresponding layers are set according to the time-frequency features of different frequency domain intervals, and fusion is performed through a fully connected layer. A weight distribution layer is set in the feature extraction model, and the attention mechanism is used to obtain the dynamic weights of the time-frequency features in different frequency domain intervals to represent the contribution of the time-frequency features; The weighted time-frequency features are compressed again through convolution and average pooling operations, and the compressed time-frequency features are dimensionality reduced and fused before being imported into the fully connected layer. The probability distribution of the target category is calculated by a normalized exponential function, the target classification result is determined according to the probability distribution, and the target classification result of the target area is combined with the corresponding spectral characteristics to generate the photoelectric measurement data acquisition result of the target area.

6. The data acquisition and processing method for photoelectric measurement according to claim 1, characterized in that: The energy distribution change corresponding to the fine target segmentation result is obtained through the multi-period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change, specifically: Extracting optimized main frequency components and energy distribution characteristics according to the optimized spectral characteristics, obtaining calibration reference points using the optimized main frequency components and energy distribution characteristics, and generating a calibration reference point set; Obtaining a fine target segmentation result through multi-period photoelectric measurement data of the target area, and performing data interpolation on the calibration reference point set according to the fine target segmentation result; The data change trend of the calibration reference point set is obtained to extract the energy distribution change, and the energy distribution change is compared with a preset calibration range. When the energy distribution change exceeds the preset calibration range, a dynamic calibration parameter is generated.

7. A data acquisition and processing system for photoelectric measurement, characterized in that: The system includes: a memory and a processor. The memory includes a data acquisition processing program for photoelectric measurement. When the data acquisition processing program is executed by the processor, the following steps are implemented: Acquire optoelectronic data of the target area, including optoelectronic sensor array data and mobile measurement device data; preprocess the acquired optoelectronic data and implement spatiotemporal registration of multi-source data to extract spatiotemporal fusion optoelectronic signals of the target area; Acquiring spectral characteristics of the measurement object based on the spatiotemporal fusion photoelectric signal, dividing a preliminary detection area using the spectral characteristics, intercepting the photoelectric signal of the detection area to perform pattern recognition to complete fine target segmentation, and optimizing the spectral characteristics; A feature extraction model is constructed based on the time-frequency analysis method. The optimized spectral features are used as the model input. The target is classified through multi-scale decomposition. The photoelectric measurement data acquisition results of the target area are generated based on the classification results and the corresponding spectral features. The energy distribution change corresponding to the fine target segmentation result is obtained through the multi-time period photoelectric measurement data of the target area, and the dynamic calibration parameters are generated according to the energy distribution change.

8. The data acquisition and processing system for photoelectric measurement according to claim 7, characterized in that: The spectral characteristics of the measurement object are obtained according to the spatiotemporal fusion photoelectric signal, and the preliminary detection area is divided using the spectral characteristics, specifically: Acquire the spatiotemporal fusion photoelectric signal of the target area, generate the corresponding spectrum image after time-frequency transformation, and perform frequency band segmentation based on the spectrum image to obtain the main frequency component and energy distribution characteristics; Performing regional interception on the spectrum image according to the energy distribution characteristics to suppress high-frequency noise, obtaining the target core frequency band by intercepting the optimized spectrum image, obtaining the time domain distribution of the spatiotemporal fusion photoelectric signal corresponding to the core frequency band, and converting the time domain distribution into a grayscale spectrum; performing sub-pixel frequency domain positioning in the grayscale spectrum to obtain frequency band edge information, iteratively optimizing the frequency band through least squares ellipse fitting based on the edge information, obtaining the energy center of the optimized ellipse, determining the frequency band width based on the energy center, and generating frequency domain features; Generate spectral features based on the main frequency component, energy distribution characteristics and frequency domain characteristics, use historical spatiotemporal fusion photoelectric signals to extract corresponding spectral features to construct a training data set, and train a convolutional neural network with the training data set to segment the measurement target; Each spectral feature is generated based on sliding window regression, and the corresponding feature area is labeled in the frequency domain to achieve preliminary division of the detection area.

9. The data acquisition and processing system for photoelectric measurement according to claim 7, characterized in that: The photoelectric signal of the detection area is intercepted and pattern recognition is performed to complete fine target segmentation, and the spectral characteristics are optimized, specifically: Acquire a preliminary detection area, perform clustering in different frequency domain intervals of the preliminary detection area according to energy density thresholds, and select interval peak points in each frequency domain interval; The energy density threshold is set by extracting the distance between the target optimization ellipse energy centers through the spectral characteristics corresponding to the target, and the distance between the peak point and the nearest frequency domain signal is calculated from high frequency to low frequency to determine whether the distance is less than the energy density threshold; If it is less than, the energy distribution information of the frequency domain interval is updated; if it is greater than, the distance difference is obtained; if the distance difference is less than the preset tolerance, the energy distribution information of the frequency domain interval is updated; The boundary range of the detection area is updated according to the energy distribution information of the frequency domain interval, the clustering of all signals is completed through iteration, the fine target segmentation is completed according to the boundary range of the updated detection area, and the spectral characteristics are optimized.

10. The data acquisition and processing system for photoelectric measurement according to claim 7, characterized in that: A feature extraction model is constructed based on the time-frequency analysis method, specifically: A feature extraction model is constructed based on wavelet transform to obtain optimized spectral features, which are then imported into the feature extraction model. The time-frequency features of different frequency domain intervals are extracted using convolution kernels. Hidden nodes of corresponding layers are set according to the time-frequency features of different frequency domain intervals, and fusion is performed through a fully connected layer. A weight distribution layer is set in the feature extraction model, and the attention mechanism is used to obtain the dynamic weights of the time-frequency features in different frequency domain intervals to represent the contribution of the time-frequency features; The weighted time-frequency features are compressed again through convolution and average pooling operations, and the compressed time-frequency features are dimensionality reduced and fused before being imported into the fully connected layer. The probability distribution of the target category is calculated by a normalized exponential function, the target classification result is determined according to the probability distribution, and the target classification result of the target area is combined with the corresponding spectral characteristics to generate the photoelectric measurement data acquisition result of the target area.

Images