DBSCAN clustering hyperspectral rock mine detection algorithm based on quantum whale swarm optimization

Abstract

The invention provides a DBSCAN clustering hyperspectral rock and mine detection algorithm based on quantum whale swarm optimization, and overcomes the limitation of the traditional technology in the aspects of high-dimensional data processing, clustering precision and noise interference. According to the method, the quantum entanglement and whale swarm algorithm are combined, DBSCAN parameters are optimized, efficient and accurate rock and ore component analysis is achieved, and manual intervention is reduced. Through a Savitzky-Golay filtering and polynomial fitting preprocessing technology, the hyperspectral data is denoised and smoothed, and main features are reserved. The neighborhood radius and the minimum point number are optimized through the whale swarm algorithm, and dependence on fixed parameters is eliminated. Features are automatically extracted in combination with a deep learning model, the calculation complexity is reduced, and the clustering accuracy is improved. And the fitness function evaluates the clustering effect, intelligently adjusts parameters, realizes high-precision rock mine detection, improves the analysis reliability, and is suitable for various scenes.

Description

Technical Field

[0001] The present invention belongs to the technical fields of spectroscopy and deep learning, and particularly relates to a DBSCAN clustering hyperspectral rock and mineral detection algorithm based on quantum whale optimization. Background Technique

[0002] Detection plays an important role in resource development, environmental protection, and future exploration, etc. Traditional sampling and analysis methods usually require bringing rock and mineral samples back to the laboratory for detailed chemical analysis. However, through spectral analysis, the difficulties and costs of actual sampling can be avoided, and the composition and characteristic information of rock and minerals can be quickly obtained. Moreover, more efficient rock and mineral detection can be carried out in extreme environments. Rock and mineral chemical analysis usually adopts methods such as X-ray fluorescence and atomic absorption spectroscopy, which require expensive equipment, reagents, and high maintenance costs. At the same time, these methods require complex sample pretreatment and long analysis time when dealing with a large number of samples, have high requirements for the skills of operators, and are difficult to meet the needs.

[0003] Cluster analysis is an important technique in data mining. It can reduce the dimension of data by classifying complex data sets, thereby simplifying the data processing and analysis process. And through unsupervised learning, the natural grouping and structure in the data can be discovered, and similar spectral data can be quickly assigned to the same category, helping researchers to more clearly identify and understand data patterns.

[0004] At present, a major challenge of clustering algorithms in the field of spectral technology is the high computational complexity when dealing with high-dimensional spectral data sets. The analysis results of clustering algorithms usually depend on the parameter settings of the algorithms. Existing methods usually choose the cumbersome process of manually selecting parameters. Selecting inappropriate parameters may lead to inaccurate results, and clustering algorithms also show certain deficiencies when dealing with problems such as data noise and cluster overlap.

[0005] Deep learning is an effective means to solve the computational complexity problem of high-dimensional spectral data sets. Through automatic feature extraction, it can effectively extract meaningful features in high-dimensional spectral data. At the same time, by training the hierarchical structure of deep learning models, the high-dimensional data is mapped to a lower-dimensional feature space through nonlinear transformation, thereby reducing the computational complexity. Using the Savitzky-Golay filtering method and polynomial fitting can effectively smooth spectral data, reduce the influence of noise on the data, maintain the main features of the data, and at the same time can better retain the details and features of the data, identify and distinguish overlapping clusters, and improve the separation degree of clusters and the accuracy of clustering.

[0006] In traditional DBSCAN, especially when dealing with high-dimensional data, the convergence speed may be slow because a large number of points need to be traversed for distance calculation. Through quantum whale optimization, using the quantum interference effect, the exploration of the clustering space can be completed in a shorter time, thereby accelerating the convergence speed of the entire clustering process and significantly reducing the time complexity of the algorithm.

[0007] The performance of the DBSCAN algorithm is limited by the selection of appropriate neighborhood radius (∈) and minimum number of points (minPts). Traditional DBSCAN clustering methods may be very sensitive to the selection of these parameters in high-dimensional space and have a large amount of computation. By using the whale algorithm to automatically select appropriate parameter values and accelerate the positioning of density regions, the efficiency of the clustering process can be effectively improved. The introduction of quantum whale optimization makes the parameter selection process not only more accurate but also faster, thus reducing the computational complexity of the DBSCAN algorithm.

[0008] The adaptive optimization during the clustering process enables the whale swarm to continuously adjust the search strategy according to the distribution characteristics of the data. With the dynamic update of the positions of individual whales in the swarm, it can more accurately adapt to the clustering needs of different data, avoid the insufficient handling of data set inconsistencies in traditional methods, and thus reduce ineffective calculations. Summary of the Invention

[0009] Object of the Invention: The object of the present invention is to provide a DBSCAN clustering hyperspectral rock and mineral detection algorithm based on quantum whale optimization to solve the computational complexity problem in high-dimensional spectral data clustering. Automatically optimize the key parameters of DBSCAN through QWOA to achieve adaptive parameter selection and improve clustering accuracy. Combine deep learning for feature extraction to reduce data dimensions and retain important information. At the same time, use the silhouette function to evaluate the clustering effect and optimize the stability and convergence speed of the algorithm. This method enhances the global search ability, avoids local optimal solutions, is efficient, accurate and adaptive, and is suitable for clustering analysis of hyperspectral data.

[0010] Technical Solution: A DBSCAN clustering hyperspectral rock and mineral detection algorithm based on quantum whale optimization includes the following steps:

[0011] S1: Preprocess the hyperspectral data: Use the Savitzky-Golay filtering method and polynomial fitting to smooth and denoise the high-dimensional spectral data, retain the key features of the data, and reduce the influence of noise. Input the denoised hyperspectral data into a deep learning model (convolutional neural network and autoencoder), extract low-dimensional features through deep learning, and achieve dimensionality reduction and feature compression of the high-dimensional spectral data, providing efficient input for subsequent clustering analysis.

[0012] S2: Explore multiple solutions in the search space using quantum increment and interference effects: Introduce quantum increment and consider interference effects to explore multiple solutions in the search space. Through the superposition and interference of quantum states, multiple solutions are searched simultaneously at the quantum level, enabling the finding of better solutions in a shorter time. This parallel search ability significantly enhances the global search ability of the traditional whale swarm algorithm and avoids falling into local optimal solutions.

[0013] S3: Quantum measurement guides the search path: The quantum measurement process causes the collapse of the quantum state, thus determining the update path of the individual. Through quantum measurement, the state of the whale individual changes according to the measurement result, thereby updating its position.

[0014] S4: Optimize DBSCAN parameters: Automatically optimize two key parameters of DBSCAN - the neighborhood radius (∈) and the minimum number of points (minPts) through the quantum whale swarm optimization algorithm. By adaptively adjusting these parameter values, the whale swarm algorithm can quickly find the density regions and distribution characteristics in the data, thus accelerating the clustering process of the DBSCAN algorithm.

[0015] S5: Evaluate the clustering effect and select the best solution: Use the silhouette coefficient to evaluate the clustering effect. Based on the fitness evaluation results, the whale swarm algorithm updates the position of each individual through techniques such as quantum entanglement and quantum measurement. This process combines the feedback of the silhouette function to ensure that the individuals converge rapidly towards the optimal solution and finally select the most suitable clustering solution.

[0016] Furthermore, the implementation details of deep learning in step S1 are as follows:

[0017] S11: Construct a convolutional neural network or an autoencoder, with the input being the preprocessed hyperspectral data;

[0018] S12: Extract the local features of the data through the convolutional layer and map the high-dimensional data to a low-dimensional space;

[0019] S13: Train the model, optimize the feature extraction process, and perform clustering analysis on the trained feature vectors.

[0020] Furthermore, the implementation details of exploring multiple solutions using quantum increment and interference effects in step S2 are as follows:

[0021] S21: In each iteration, the quantum whale swarm algorithm superimposes multiple positions in the search space and enhances the probability of the optimal solution according to the quantum interference effect. Through the parallelism of quantum increment, the algorithm can quickly explore the region of the optimal solution or near-optimal solution;

[0022] S22: The interference effect in quantum computing exerts an interference effect between different solutions by affecting the probability amplitude of the solutions, helping the algorithm focus on the most promising solution space region and avoiding ineffective searches.

[0023] Furthermore, S3: Specifically, it includes the following key points:

[0024] S31: Quantum measurement leads to the collapse of the quantum state, and the individual quantum state becomes a definite value to determine its new search position. The search path of each individual in the whale group is determined by the quantum measurement result, enabling the individual to quickly converge towards the global optimal solution.

[0025] S32: Dynamically update the positions of whale individuals. The position of each individual will be updated according to the quantum measurement result to guide the search direction. The quantum measurement process allows the individual to dynamically adjust its search position, enabling the whale group to effectively search the entire space.

[0026] Furthermore, the implementation details in step S4 are as follows:

[0027] The quantum whale group optimization algorithm uses the quantum superposition and entanglement technologies in quantum computing to explore different combinations of ∈ and minPts. Each time, the parameter values are updated according to the fitness evaluation result (silhouette coefficient), thus accelerating the clustering process.

[0028] Furthermore, the clustering evaluation in step S5 is implemented as follows:

[0029] S41: Quantify the clustering effect by calculating the silhouette coefficient and identify the best clustering scheme;

[0030] S42: The whale group algorithm updates the position of each individual according to the fitness evaluation result to ensure that the algorithm quickly converges to the optimal solution. Integrating technologies such as quantum entanglement and quantum measurement can further accelerate the update of individuals and ensure that the optimal clustering scheme is finally selected.

[0031] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0032] 1. The prior art analyzes the rock and ore components through the coordinates of rock and ore collection and three-dimensional models, but the processing process is time-consuming and complex. The present invention introduces deep learning and automatic feature extraction technologies, which can effectively reduce the processing complexity of the original high-dimensional spectral data and achieve rapid analysis of rock and ore samples. Especially when processing large-scale spectral data sets, the present invention demonstrates better computational efficiency and can greatly shorten the data processing time.

[0033] 2. Most of the prior art relies on preset models for rock and ore component analysis and is easily interfered by noise, resulting in limited result accuracy. The present invention combines Savitzky-Golay filtering, polynomial fitting, and deep learning models to automatically remove noise and retain the key features of the spectrum. This automated processing flow can not only improve the clarity of the data but also more accurately identify and distinguish different types of rock and ore in subsequent clustering analysis, significantly improving the accuracy of component analysis.

[0034] 3. The existing technology can enhance the global search ability. The quantum increment and interference effects introduce the advantages of quantum computing, enabling individual whales in the whale swarm to search for multiple solutions in parallel at the quantum level, which can greatly improve the search efficiency and global search ability. This technology can avoid the limitation that traditional optimization methods are prone to falling into local optimal solutions, enabling the algorithm to find the optimal solution in a shorter time.

[0035] 4. Traditional technologies usually rely on fixed clustering algorithm parameters that need to be manually adjusted, which increases the difficulty and uncertainty of parameter tuning. By introducing an adaptive parameter adjustment mechanism, the quantum whale swarm optimization algorithm automatically optimizes two key parameters of DBSCAN - the neighborhood radius (∈) and the minimum number of points (minPts), and can adjust these parameters according to the density characteristics of the data. This adaptive optimization process can accelerate the clustering speed of the DBSCAN algorithm and improve the accuracy and robustness of clustering.

[0036] 5. Traditional technologies rely on the methods of collecting coordinates and three-dimensional models for visualizing the distribution of rock and mineral components, which are cumbersome and costly. Through intelligent spectral data analysis and deep learning models, the present invention can automatically analyze the rock and mineral components in complex environments (such as deep sea, space) without a large number of samplings, which can significantly reduce the sample collection cost. At the same time, the intelligent analysis method of the present invention improves the efficiency of rock and mineral exploration, can be applied to a wider range of scenarios, and promotes the intelligence and automation of rock and mineral analysis.

[0037] 6. Compared with traditional heuristic-based clustering algorithms (such as K-means, DBSCAN, etc.), the present invention combines quantum computing and the whale swarm optimization algorithm, and can achieve parallel processing during the search process, significantly improving the computing efficiency. Traditional methods often face problems such as excessive consumption of computing resources or slow search processes when dealing with high-dimensional data, while the present invention effectively overcomes these problems through quantum increment and interference effects, providing a faster computing speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] To more clearly illustrate the embodiments of the present invention and its design, the drawings required for this embodiment will be briefly introduced below. The drawings described below are only partial embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 Flowchart of hyperspectral rock and mineral analysis based on improved clustering

[0040] Figure 2 Deep learning network architecture diagram DETAILED DESCRIPTION OF THE INVENTION

[0041] Now, by referring to exemplary embodiments, the objects and functions of the present invention and the methods for achieving these objects and functions will be elucidated. However, the present invention is not limited to the exemplary embodiments disclosed below; it can be implemented in different forms. The essence of the specification is merely to assist those skilled in the relevant art in comprehensively understanding the specific details of the present invention.

[0042] Embodiment 1

[0043] The data used in this embodiment is the Raman spectroscopy dataset of rock and ore samples R080016 - R080100: widely scanned and the samples are not oriented, including 6440 items of data, and each item of data is sampled approximately 1200 times from a wavelength of 250 cm -1 to a wavelength of 3000 cm -1 The present invention provides a DBSCAN clustering hyperspectral rock and ore detection algorithm based on quantum whale optimization. The specific process is as Figure 1 shown, including the following steps:

[0044] Step A: Obtain the spectral data for rock and ore detection, and preprocess the obtained spectral data for rock and ore detection to optimize subsequent deep learning and clustering processing, and then proceed to Step B.

[0045] In practical applications, the above Step A is specifically executed as Steps A1 to A3 as follows:

[0046] Step A1: Import the existing large dataset of rock and ore spectral detection, obtain the intensity values of different compounds from wavelengths 250 - 3000, and use "StandardScaler" to standardize the spectral vectors of each band so that the amplitude mean is 0 and the variance is 1 on all bands, generating a normalized large dataset of rock and ore spectral detection;

[0047] Step A2: For the normalized large dataset of rock and ore spectral detection, apply the Savitzky - Golay filtering method to smooth the data through local polynomial fitting. Fit a polynomial in the neighborhood of x i Using the smoothing coefficient matrix, given a window size of 11 and setting the polynomial order to 2, obtain the smoothing coefficients of the filter through the least - squares method. The coefficient matrix M of the filter is solved through the least - squares method and is used to smooth the original data x to the new data, generating a smoothed large dataset of rock and ore spectral detection, and then proceed to Step A3;

[0048] where, c j is the coefficient obtained from the filter design, x i+j is the value in the original data, and y i is the smoothed data;

[0049] Step A3: For the smoothed large dataset of rock and ore spectral detection, use polynomial fitting to further process the data of Savitzky-Golay filtering. Apply polynomial fitting to each feature to generate a denoised large dataset of rock and ore spectral detection, completing the preprocessing of the data.

[0050] Step B: For the preprocessed large dataset of rock and ore spectral detection, train a deep learning model (convolutional neural network and autoencoder) to reduce the dimension of high-dimensional spectral data through deep learning. As Figure 2 shown, we input the tabular image of rock and ore spectral data. Through the first convolutional layer, 32 different convolutional filters are used to perform convolutional operations on our tabular image. Based on the feature map of the first convolutional layer, through the second convolutional layer, 64 different convolutional filters are used to extract higher-level features. Based on the second convolutional layer, 128 filters are used on the third convolutional layer to extract more complex data. When passing through each convolutional layer, a 2×2 pooling window is used to reduce the spatial dimension of the feature map and maintain the main features of the rock and ore spectral image. After passing through the convolutional layer and the pooling layer, the feature map of the rock and ore spectral data is flattened into a one-dimensional vector, and the data is processed through a fully connected layer of 256 dimensions. Finally, the softmax activation function is used through the output layer of 10 dimensions to achieve the purpose of extracting low-dimensional features, providing efficient input for subsequent clustering analysis, and then entering Step C.

[0051] Step C: Perform quantum whale swarm optimization on the dimension-reduced data and enter Step C1.

[0052] Step C1: Initialize the preliminary whale swarm optimization. The whale swarm algorithm initializes P = 20 individuals, and each individual is represented as a solution p = (∈, minPts), where ∈ is the neighborhood radius of DBSCAN and minPts is the minimum number of points. Assume the initial neighborhood radius ∈ i ranges from [0.1, 1.0], and the minimum number of points minPts i ranges from [5, 100], and enter Step C2. pi = (∈i, minPtsi) for i = 1, 2, …, P

[0053] Step C2: Perform quantum increment and interference effects. The solutions of each whale individual are superimposed at the quantum level, and the states of all individuals exist at multiple positions simultaneously. The probability of the optimal solution is enhanced through quantum interference. Assume the quantum superposition state of a certain individual is: |ψ i > = α|p i > + β|p j > where, |p i > and |pj > represents the states at different positions, where α and β are probability amplitudes, and proceed to step C3.

[0054] Step C3: Perform quantum measurement and position update. After the quantum measurement, the quantum state of the whale individual will collapse, resulting in the update of the position of its solution, and proceed to step D. pi = measure(ψ i )

[0055] Step D: Use the silhouette function to evaluate the clustering result. The closer the value is to 1, the better the data partitioning is, and the closer it is to -1, the more it indicates being misassigned to the current cluster. Calculate the silhouette coefficient: According to the fitness evaluation result, use quantum entanglement and quantum measurement to update the position of the whale individual, and re-enter step C. Through multiple quantum measurement cycles, continuously optimize the solution until it finally converges to the optimal solution.

[0056] The above step D is specifically as follows in the embodiment:

[0057] In the embodiment, perform DBSCAN clustering using the optimized parameters ∈ and minPts. Select ∈ = 0.5 and minPts = 10 as the optimized parameters. Use the silhouette coefficient to evaluate the clustering effect, and the silhouette coefficient s = 0.75. Through multiple iterations, finally select the clustering scheme with the highest silhouette coefficient as the final scheme.

[0058] The above-described embodiments are only the preferred specific embodiments of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of the technical solutions that can be obviously obtained by those skilled in the art within the technical scope disclosed by the present invention shall fall within the protection scope of the present invention.

Claims

1. A hyperspectral rock and mineral detection method based on quantum whale swarm optimization for DBSCAN clustering, specifically including the following steps: S1 Preprocess the hyperspectral data, use the Savitzky-Golay filtering method and the smoothing algorithm to denoise, and perform smoothing processing on the high-dimensional spectral data through polynomial fitting to retain the main features of the spectral data; S2 Use a deep learning model (including a convolutional neural network and an autoencoder) to extract the low-dimensional features of the hyperspectral data, reduce the data dimension through automatic feature extraction, and compress the features; S3 Use quantum increment and interference effects to explore multiple solutions in the search space, generate a quantum whale swarm optimization algorithm, and make the exploration of efficient whale individuals in the search space closer through quantum entanglement; S4 The "encircling" and "searching" processes of the whale swarm algorithm use quantum measurement to update the position of each individual, and use the collapse decision of the quantum state generated by the quantum measurement process to optimize the update step of the decision of each whale swarm individual, guiding the search path of the whale swarm algorithm; S5 Use the whale swarm algorithm to find the appropriate neighborhood radius (∈) and the minimum number of points (minPts), select a set of parameter values, perform iterations using DBSCAN, and use the whale swarm algorithm to help find the possible density regions and distribution characteristics in the data, accelerating the positioning of the DBSAN density region; S6 Evaluate the key results according to the fitness function, update the position (parameter value) of each individual, select the best clustering scheme and output the clustering label.

2. The DBSCAN clustering hyperspectral rock and mineral detection method based on quantum whale optimization according to claim 1, wherein, The deep learning model in step S2 is a convolutional neural network or an autoencoder, which extracts the local features of the data through the convolutional layer, and gradually extracts the multi-level features of the spectral data through the structure of the convolutional layer, pooling layer and fully connected layer to achieve efficient feature compression and map the high-dimensional data to the low-dimensional space.

3. The DBSCAN clustering hyperspectral rock and mineral detection method based on quantum whale optimization according to claim 2, wherein, The quantum increment and correlation effects in step S3 explore multiple solutions in the search space and generate a quantum whale swarm optimization algorithm. Quantum bits are correlated with whale swarm individuals, and the individual state is represented as a quantum state, thereby accelerating the close cooperation between whale swarm individuals through quantum entanglement, optimizing the search path, and enhancing the global exploration ability.

4. The DBSCAN clustering hyperspectral rock and mineral detection method based on quantum whale optimization according to claim 3, wherein, In the classical implementation of the whale swarm algorithm, the movement of each whale individual fish is usually based on the adaptation of the objective function. In step S4, quantum computing can provide efficient search ability and more global exploration ability. The superposition state of quantum bits is used to characterize the state of whale swarm individuals. When quantum measurement is performed, the system will collapse to a specific ground state. This process transforms the global search advantage of quantum computing in the solution space into the generation mechanism of individual position decision-making in the swarm intelligence algorithm, effectively integrating the dynamic characteristics of quantum parallelism and swarm iterative optimization.

5. The DBSCAN clustering hyperspectral rock and mineral detection method based on quantum whale optimization according to claim 3, wherein, The quantum whale swarm optimization in step S5 can automate and accelerate the selection of DBSCAN parameters. The DBSCAN algorithm calculates the density of data points, identifies noise points and discovers clusters of arbitrary shapes, and uses a distance metric method to judge the membership of points.

6. The DBSCAN clustering hyperspectral rock and mineral detection method based on quantum whale optimization according to claim 4, wherein, The clustering effect evaluation in step 6 quantifies the clustering result through the silhouette coefficient, and selects the clustering scheme with the silhouette coefficient value closest to 1.

Images