Crude oil exploration method based on image detection

Abstract

The present invention provides a crude oil exploration method based on image detection, relating to the technical field of crude oil exploration. The method comprises collecting various data; constructing an enhanced anchor-free detection model optimized based on the Giant Armadillo algorithm, inputting historical crude oil exploration data, and outputting crude oil feature image data. The geological structure image data is analyzed to obtain key feature image data, and surface image features are extracted from the surface image data. The key feature image data and the surface image features are combined to obtain complete feature image data, which is then compared with the crude oil feature image data to determine whether crude oil is present in the target area. By optimizing the anchor-free detection model using the Giant Armadillo algorithm, the present invention improves the efficiency and accuracy of crude oil exploration and enhances the model's adaptability and generalization capabilities. Furthermore, the complete feature image data obtained through feature extraction and fusion provides a deeper understanding of the geological conditions of the target area.

Description

Technical Field

[0001] The present invention relates to the technical field of crude oil exploration, and in particular to a crude oil exploration method based on image detection. Background Art

[0002] In the field of crude oil exploration, image detection methods have gradually become an important exploration tool because they can intuitively obtain and analyze geological information. With the continuous development of exploration technology, higher requirements are placed on the accuracy, efficiency and reliability of image detection methods.

[0003] Currently, traditional image-based crude oil exploration methods mostly rely on object detection models to identify features related to crude oil. However, these detection models generally have certain limitations. Many models use an anchor-based detection approach, which requires pre-setting a large number of anchors of varying sizes and proportions to cover possible targets. This significantly increases the computational load, making model training and inference complex and time-consuming. Furthermore, the anchor setting needs to be adjusted based on the specific exploration scenario. Finding an optimal set of anchor parameters for targets of varying scales and shapes is difficult, making the model less adaptable to the complex and varied crude oil exploration imagery, prone to missed or false detections, and impacting the accuracy and efficiency of crude oil exploration.

[0004] Furthermore, existing technologies for feature extraction and fusion of image data also suffer from inefficiencies. Geological structure and surface image data contain a wealth of information relevant to crude oil exploration, but traditional feature extraction methods often fail to fully tap into this information. During the feature fusion stage, there is a lack of accurate assessment of feature importance and a rational fusion strategy. This inappropriate feature fusion approach prevents the fused feature data from fully leveraging the strengths of each feature, making it difficult to provide comprehensive and accurate information for crude oil exploration decisions, thereby reducing the reliability of exploration results. Summary of the Invention

[0005] The present invention provides a crude oil exploration method based on image detection, which is used to solve the defects of the existing technology, such as the limitations of the detection model and the unreasonable extraction and fusion of image data features.

[0006] In one aspect, the present invention provides a crude oil exploration method based on image detection, comprising:

[0007] Collect geological structure image data and historical crude oil exploration data within the target area, and collect surface image data at preset time intervals.

[0008] An anchor-free detection model is constructed and optimized using the giant armadillo algorithm to obtain an enhanced anchor-free detection model. Historical crude oil exploration data is input and crude oil feature image data is output.

[0009] The geological structure image data is analyzed using the feature analysis method to obtain key feature image data. The surface image data is extracted based on the multi-scale geometric analysis feature method to obtain surface image features. The key feature image data is combined with the surface image features based on the feature fusion method to obtain complete feature image data.

[0010] The complete feature image data is compared with the crude oil feature image data to determine whether crude oil exists in the target area.

[0011] According to the crude oil exploration method based on image detection provided by the present invention, the steps of constructing an anchor-free frame detection model include:

[0012] The basic features of images in crude oil exploration data are extracted and learned through residual blocks to build a network structure.

[0013] In the output layer of the network structure, a heat map is obtained through convolution operation, and the center point of the network structure is predicted. At the position of the center point, the width and height of the network structure are predicted through the convolution layer, and the key points of the target are determined based on the center point, width and height.

[0014] Focal-Loss is used to train the heat map and calculate the center point loss.

[0015] L1-Loss is used to calculate the difference between the actual value and the preset value to obtain the size loss.

[0016] The weighted sum of center point loss and size loss is used as the total loss function.

[0017] According to the crude oil exploration method based on image detection provided by the present invention, the steps of obtaining the enhanced anchor-free frame detection model include:

[0018] The giant armadillo population size is set, and each individual armadillo represents a set of parameters for the anchor-free box detection model. The parameter value of each individual armadillo is randomly initialized within a preset value range.

[0019] For each armadillo individual, a new digging individual is generated through digging behavior.

[0020] For each armadillo individual, a new escape individual is generated by escaping after sensing a threat.

[0021] Calculate the fitness values ​​of the initial armadillo individual, the digging individual, and the escaping individual, and select the armadillo individual with the highest fitness value as the individual for the next iteration.

[0022] When the preset number of iterations is reached, the parameter value represented by the armadillo individual with the highest fitness value is selected and applied to the anchor-free detection model to obtain an enhanced anchor-free detection model.

[0023] According to the crude oil exploration method based on image detection provided by the present invention, the step of outputting crude oil characteristic image data includes:

[0024] The historical crude oil exploration data is converted into the input requirements that meet the enhanced anchor-free detection model, and normalized to obtain the exploration processing data.

[0025] The exploration and processing data is input into the enhanced anchor-free box detection model, forward propagation calculation is performed through each layer, and feature extraction is performed to obtain feature map data.

[0026] The key points and sizes of the feature map data are predicted, the features of different layers are fused, and the predictions that do not meet the preset requirements are removed to obtain the prediction results.

[0027] The prediction results are converted into image form and integrated together to obtain crude oil characteristic image data.

[0028] According to the crude oil exploration method based on image detection provided by the present invention, the step of obtaining characteristic map data includes:

[0029] In the first layer of the network structure, a convolution operation is performed on the exploration processing data to obtain the convolution result, and an activation function is used to perform nonlinear changes to obtain the output feature map.

[0030] In the middle layer of the network structure, convolution operations and activation function operations are repeated. When the preset number of times is reached, high-level abstract features are extracted.

[0031] In the last layer of the network structure, the output feature map and high-level abstract features are output as feature map data.

[0032] According to the crude oil exploration method based on image detection provided by the present invention, the step of obtaining key feature image data includes:

[0033] The geological structure image data was denoised using the Gaussian filtering algorithm, and the contrast was enhanced using the histogram equalization method.

[0034] The processed geological structure image data is subjected to edge detection, texture analysis and morphological feature extraction to obtain multiple target geological structure features, and the correlation between the structural features of each target address is calculated to obtain the correlation relationship.

[0035] According to the correlation relationship and analysis purpose, each target geological target structure is screened to obtain key features.

[0036] The key features are visualized, converted into image form, and integrated to obtain key feature image data.

[0037] According to the crude oil exploration method based on image detection provided by the present invention, the step of obtaining the correlation relationship includes:

[0038] Multiple target geological structure characteristics are expressed as vectors, and the Pearson correlation coefficient is used to calculate the correlation between each two target geological structure characteristic vectors to obtain a correlation matrix. The correlation relationship is obtained by analyzing the correlation matrix.

[0039] According to the crude oil exploration method based on image detection provided by the present invention, the step of extracting surface image features includes:

[0040] The color image in the surface image data is converted into a grayscale image, and denoising is performed to obtain the surface image processing data.

[0041] Curvelet is used to perform multi-scale decomposition on surface image processing data, and directional decomposition is performed at each scale to obtain Curvelet coefficients.

[0042] Statistical analysis is performed on the Curvelet coefficients of each scale and direction to extract relevant features. Curvelet coefficients smaller than a preset threshold are removed, and geometric structure features are extracted based on the distribution and changes of the Curvelet coefficients.

[0043] The surface image features are obtained by fusing the relevant features extracted at different scales and the geometric structure features extracted in different directions.

[0044] According to the crude oil exploration method based on image detection provided by the present invention, the steps of obtaining complete characteristic image data include:

[0045] The key feature image data and surface image features are standardized, and the maximum pooling method is used to make the dimensions consistent.

[0046] The preset weights of key feature image data and surface image features are obtained according to the principal component analysis method.

[0047] Different weights are assigned to key feature image data and surface image features according to preset weights, and weighted addition is performed to obtain complete feature image data.

[0048] According to the crude oil exploration method based on image detection provided by the present invention, the step of obtaining the preset weight includes:

[0049] The key feature image data and the surface image features are combined into a feature matrix, and then normalized to obtain a standard feature matrix.

[0050] Calculate the covariance matrix of the standard eigenvalue matrix and perform eigendecomposition to obtain multiple eigenvalues ​​and corresponding eigenvectors.

[0051] Arrange multiple eigenvalues ​​in descending order, select a preset number of eigenvalues ​​and corresponding eigenvectors to form a principal component matrix, and calculate the contribution weights of each key feature image data and each surface image feature in the principal component matrix to obtain the preset weights.

[0052] The crude oil exploration method based on image detection provided by the present invention optimizes the anchor-free detection model by using the giant armadillo algorithm, thereby solving the problem that traditional target detection models may be difficult to achieve optimal performance due to their dependence on anchor frames and parameter settings. This has the beneficial effects of improving the efficiency and accuracy of crude oil exploration and enhancing the adaptability and generalization ability of the model.

[0053] The crude oil exploration method based on image detection provided by the present invention solves the problem of difficulty in data feature extraction and fusion by performing detailed feature extraction and fusion of geological structure image data and surface image data, and achieves the beneficial effect of providing a more comprehensive reference for geological experts and exploration personnel, helping to gain an in-depth understanding of the geological conditions of the target area and providing a more powerful basis for crude oil exploration decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 This is one of the flow diagrams of the crude oil exploration method based on image detection provided by an embodiment of the present invention;

[0056] Figure 2 This is the second flow chart of the Internet of Things-based park inspection device provided by an embodiment of the present invention;

[0057] Figure 3 This is the third flow chart of the Internet of Things-based park inspection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0059] The following combination Figure 1-Figure 3 The crude oil exploration method based on image detection of the present invention is described.

[0060] like Figure 1 As shown, the crude oil exploration method based on image detection provided by the embodiment of the present invention includes:

[0061] Geological structure image data and historical crude oil exploration data within the target area are collected, and surface image data is acquired at preset intervals. Using geological radar technology, high-frequency electromagnetic waves are transmitted underground, and the reflected signals are received and processed. High-resolution geological structure images are generated through signal analysis. These images clearly demonstrate the stratigraphic structure, rock types, and possible geological features such as faults and cavities at different depths. Within the target area, geological radar detection points are arranged at a predetermined grid density to ensure full coverage, acquire comprehensive geological structure information, and thus obtain geological structure image data.

[0062] From drilling logs, well logging data, core analysis reports, oil test and production test data, etc. Through the collation and analysis of historical data, we can understand the distribution pattern of crude oil in the target area, the physical properties of the oil layer, the chemical composition of crude oil and other important information to obtain historical crude oil exploration data.

[0063] Collecting surface image data at preset intervals is an important way to monitor surface changes in a target area in real time. Using drones equipped with high-resolution cameras, aerial photography of the target area is performed at preset intervals once a week.

[0064] An anchor-free detection model is constructed and optimized using the giant armadillo algorithm to obtain an enhanced anchor-free detection model. Historical crude oil exploration data is input and crude oil feature image data is output.

[0065] The steps to build an anchor-free detection model include:

[0066] Extract the basic features of the image in the crude oil exploration data, and learn the basic features through the residual block to build a network structure. The formula is expressed as:

[0067]

[0068] in, It is The input of the layer, It is The output of the layer, is the weight, It's about input and weights The residual function of .

[0069] At the output layer of the network structure, a heat map is obtained through convolution operation, and the center point of the network structure is predicted. At the location of the center point, the width and height of the network structure are predicted through the convolution layer. The key points of the target are determined based on the center point, width and height. The formula for the size of the heat map is expressed as:

[0070]

[0071] Where, is the height of the heat map, is the width of the heatmap, is the number of categories of the heat map.

[0072] Focal-Loss is used to train the heat map and calculate the center point loss. The formula is:

[0073]

[0074] Where, is the number of objects in the image, is the predicted heat map at location For category The probability of It is a real heat map. and is a hyperparameter.

[0075] Use L1-Loss to calculate the gap between the actual value and the preset value to get the size loss. The formula is:

[0076]

[0077] Where, It is Target preset values, It is The actual value of a target.

[0078] The weighted sum of the center point loss and the size loss is used as the total loss function, and the formula is expressed as:

[0079]

[0080] Where, is a hyperparameter used to balance the size loss and the center point loss, is the total loss function.

[0081] like Figure 2 As shown in Figure 2, the steps to obtain the enhanced anchor-free detection model include:

[0082] The giant armadillo population size is set, and each individual armadillo represents a set of parameters for the anchor-free box detection model. The parameter value of each individual armadillo is randomly initialized within a preset value range.

[0083] For each armadillo individual, a new mining individual is generated through mining behavior. The formula is expressed as:

[0084]

[0085] Where, is the parameter that controls the step size, and is a random number between [0,1], is the individual with the best fitness in the current population. It is Armadillo individuals, It is a new excavation individual.

[0086] For each armadillo individual, a new escape individual is generated by escaping through sensing the threat. The formula is expressed as:

[0087]

[0088] Where, Is the new escape individual, is an individual randomly selected from the population, is a random number between [0,1].

[0089] Calculate the fitness values ​​of the initial armadillo individual, the digging individual, and the escaping individual, and select the armadillo individual with the highest fitness value as the individual for the next iteration.

[0090] When the preset number of iterations is reached, the parameter value represented by the armadillo individual with the highest fitness value is selected and applied to the anchor-free detection model to obtain an enhanced anchor-free detection model.

[0091] The steps of outputting crude oil characteristic image data include:

[0092] The historical crude oil exploration data is converted into the input requirements that meet the enhanced anchor-free detection model, and normalized to obtain the exploration processing data.

[0093] The exploration and processing data is input into the enhanced anchor-free box detection model, forward propagation calculation is performed through each layer, and feature extraction is performed to obtain feature map data.

[0094] The steps to obtain feature map data include:

[0095] In the first layer of the network structure, a convolution operation is performed on the exploration and processing data to obtain the convolution result. The activation function is used to perform nonlinear transformations to produce the output feature map. The convolution operation extracts local features of the data. The convolution kernel is element-wise multiplied with the corresponding region of the input data and the sum is calculated to obtain the convolution result.

[0096] In the middle layer of the network structure, convolution operations and activation function operations are repeated. When the preset number of times is reached, high-level abstract features are extracted.

[0097] In the last layer of the network structure, the output feature map and high-level abstract features are output as feature map data.

[0098] The key points and sizes of the feature map data are predicted, the features of different layers are fused, and the predictions that do not meet the preset requirements are removed to obtain the prediction results.

[0099] The prediction results are converted into images and integrated to generate crude oil characteristic image data. This crude oil characteristic image data can intuitively display the model's detection results of potential crude oil characteristics in historical crude oil exploration data, including information such as the crude oil target location, size, geological structure, and surrounding environmental topography.

[0100] The geological structure image data is analyzed using the feature analysis method to obtain key feature image data. The surface image data is extracted based on the multi-scale geometric analysis feature method to obtain surface image features. The key feature image data is combined with the surface image features based on the feature fusion method to obtain complete feature image data.

[0101] The steps of obtaining key feature image data include:

[0102] The geological structure image data was denoised using a Gaussian filter, and contrast was enhanced using histogram equalization. The Gaussian filter effectively removes Gaussian noise and smoothes the image by performing a weighted average of image pixels and their neighborhoods, determining weights based on a Gaussian distribution. Histogram equalization redistributes pixel values ​​to create a more uniform histogram distribution, thereby enhancing overall image contrast and making geological structure details more apparent.

[0103] The processed geological structure image data is subjected to edge detection, texture analysis and morphological feature extraction to obtain multiple target geological structure features, and the correlation between the structural features of each target address is calculated to obtain the correlation relationship.

[0104] The steps to obtain the correlation relationship include:

[0105] Multiple target geological structure characteristics are expressed as vectors, and the Pearson correlation coefficient is used to calculate the correlation between each two target geological structure characteristic vectors to obtain a correlation matrix. The correlation relationship is obtained by analyzing the correlation matrix.

[0106] Edge detection: The edges of geological structures, such as stratum boundaries and fault lines, are key features. The Canny operator (Canny edge detection operator) is used to detect accurate and continuous edges by calculating the gradient magnitude and direction of the image and performing non-maximum suppression and double thresholding.

[0107] Texture Analysis: Different geological structures have distinct texture characteristics, such as the texture differences between sandstone and shale. By counting the frequency of pixel pairs with different grayscale values ​​in an image and calculating texture characteristic parameters such as contrast, correlation, energy, and entropy, we can describe the texture characteristics of geological structures.

[0108] Morphological feature extraction: Utilizing mathematical morphology methods such as dilation, erosion, opening, and closing, we extract the morphological features of geological structures. For example, dilation can expand connected regions within a geological structure, while erosion can remove burrs from edges. Opening and closing operations can be used to remove small noise blocks and fill small holes, respectively, thereby highlighting the overall morphological characteristics of the geological structure.

[0109] According to the correlation relationship and analysis purpose, each target geological target structure is screened to obtain key features.

[0110] The key features are visualized, converted into image form, and integrated to obtain key feature image data.

[0111] The steps of extracting surface image features include:

[0112] The color image in the surface image data is converted into a grayscale image, and denoising is performed to obtain the surface image processing data.

[0113] Curvelet is used to perform multi-scale decomposition on surface image processing data. Directional decomposition is performed at each scale to obtain Curvelet coefficients. The image in the surface image processing data is decomposed into subbands of different scales, each representing image information at a different scale. Large-scale subbands contain low-frequency information of the image, reflecting the overall structure and contours. Small-scale subbands contain high-frequency information of the image, reflecting image details and edges. Geometric structure information in different directions is captured at each scale. Directional decomposition yields multiple directional subbands, each corresponding to image features in a specific direction.

[0114] Statistical analysis is performed on the Curvelet coefficients of each scale and direction to extract relevant features. Curvelet coefficients smaller than a preset threshold are removed, and geometric structure features are extracted based on the distribution and changes of the Curvelet coefficients.

[0115] The surface image features are obtained by fusing the relevant features extracted at different scales and the geometric structure features extracted in different directions. The surface image features can include color features, texture features, shape features, spatial features, and spectral features.

[0116] like Figure 3 As shown, the steps of combining to obtain complete feature image data include:

[0117] The key feature image data and surface image features are standardized, and the maximum pooling method is used to make the dimensions consistent.

[0118] The preset weights of key feature image data and surface image features are obtained according to the principal component analysis method.

[0119] Different weights are assigned to key feature image data and surface image features according to preset weights, and weighted addition is performed to obtain complete feature image data.

[0120] The steps to obtain the preset weights include:

[0121] The key feature image data and the surface image features are combined into a feature matrix, and then normalized to obtain a standard feature matrix.

[0122] Calculate the covariance matrix of the standard eigenvalue matrix and perform eigendecomposition to obtain multiple eigenvalues ​​and corresponding eigenvectors.

[0123] Arrange multiple eigenvalues ​​in descending order, select a preset number of eigenvalues ​​and corresponding eigenvectors to form a principal component matrix, and calculate the contribution weights of each key feature image data and each surface image feature in the principal component matrix to obtain the preset weights.

[0124] The complete feature image data is compared with the crude oil feature image data to determine the presence of crude oil in the target area. The location and size of potential crude oil targets, as shown in the crude oil feature image data, are then matched against features such as geological structure and surface morphology in the complete feature image data. A quantitative comparison is performed on various feature parameters extracted from the crude oil feature image data and the complete feature image data, such as contrast and correlation in texture features, and perimeter and area in shape features. The degree of difference between these feature parameters is calculated to assess the degree of match between the two. Based on these comparison results, a final conclusion is drawn regarding the presence of crude oil in the target area. If multiple features and parameters show signs of crude oil during the comparison, and the geological structure correlation supports this conclusion, the likelihood of crude oil in the target area is high. Conversely, if the comparison results are inconsistent or lack key supporting evidence, further exploration and research are necessary to confirm the presence of crude oil.

[0125] Example 1: 1. Data collection.

[0126] Geological structure image data: Geological structure image data with a resolution of 500×500 pixels in the target area was obtained through geological radar and other equipment. A total of 100 images were collected, which reflect the geological structure at different depths underground.

[0127] Historical crude oil exploration data: We collected crude oil exploration data from the target area over the past 10 years, including previous drilling records, logging data, and image data of some marked oil-bearing areas, and compiled 80 sets of valid data records.

[0128] Surface image data: Surface images were acquired by drone photography at a preset time interval of once a week, with a resolution of 800 × 800 pixels. A total of 50 color images were collected.

[0129] 2. Build and optimize the detection model.

[0130] Building an anchor-free detection model: A basic anchor-free detection model is constructed using a residual network structure. This model extracts basic features from images in crude oil exploration data and learns these features using residual blocks. The input image data is assumed to have dimensions of 3×224×224 (number of channels × height × width). After processing with multiple layers of residual blocks, a convolution operation is performed at the output layer to produce a heat map of size 14×14×10 (height × width × number of categories, with 10 categories used to distinguish different feature types). The model then predicts the center point of the network structure, as well as the width and height at that center point, to identify the key points of the target.

[0131] Optimization model: The giant armadillo population size was set to 50. Each individual armadillo represented a set of anchor-free detection model parameters. The parameter values ​​for each individual armadillo were randomly initialized within a preset range. New individuals were generated through mining and evasion behaviors, and the fitness values ​​of the initial armadillo individuals, mining individuals, and evasion individuals were calculated. After 100 preset iterations, the parameter values ​​representing the armadillo individuals with the highest fitness were applied to the anchor-free detection model, resulting in an enhanced anchor-free detection model. Before optimization, the model achieved a mAP of 0.65 on the validation set, which improved to 0.78 after optimization.

[0132] 3. Data feature extraction

[0133] Geological structure image data:

[0134] The 100 geological structure image data were denoised using the Gaussian filtering algorithm (with the standard deviation set to 1.5), and then the contrast was enhanced using the histogram equalization method.

[0135] The Canny operator was used for edge detection, and obvious edge features such as stratum boundaries and fault lines were detected in 80 images. The GLCM was used to calculate the texture feature parameters, and the average contrast was 0.35, the average correlation was 0.68, the average energy was 0.22, and the average entropy was 1.85. Morphological features were extracted through morphological operations such as dilation and erosion, and closed or semi-closed morphological structures that are conducive to crude oil storage were found in 60 images.

[0136] Multiple target geological structural features were represented as vectors, and the Pearson correlation coefficient was used to calculate the correlation between each feature to obtain a correlation matrix. Based on the correlation relationship and analysis objectives, 30 key features were screened, visualized, and integrated to obtain key feature image data.

[0137] Surface image data:

[0138] The 50 color surface images were converted into grayscale images and denoised using median filtering to obtain surface image processing data.

[0139] The surface image processing data was decomposed using Curvelet at multiple scales (with a set number of scales of 4) and then decomposed at each scale in a directional manner (with a set number of directions of 8). Curvelet coefficients were obtained. Statistical analysis was performed on the Curvelet coefficients at each scale and direction, removing coefficients below a preset threshold (set to 0.05). Geometric structural features were then extracted based on the distribution and variation of the coefficients. Features extracted at different scales and directions were fused to obtain surface image features. For example, surface texture variations potentially associated with crude oil seepage were found in 35 images.

[0140] 4. Feature Fusion

[0141] The key feature image data and surface image features are standardized, and the maximum pooling method is used to make the dimensions consistent so that their dimensions become 1×100 (assuming the feature dimensions obtained after processing before fusion).

[0142] Using principal component analysis, the two were combined into a single feature matrix and normalized to obtain a standard feature matrix. The covariance matrix of the standard feature matrix was calculated and eigendecomposition was performed to obtain multiple eigenvalues ​​and corresponding eigenvectors. The eigenvalues ​​were arranged from largest to smallest, and the top 10 eigenvalues ​​and corresponding eigenvectors were selected to form the principal component matrix. The preset weights for the key feature image data and the surface image features were calculated to be 0.6 and 0.4, respectively.

[0143] Assign weights to key feature image data and surface image features according to preset weights, and perform weighted addition to obtain complete feature image data.

[0144] 5. Crude oil characteristics judgment

[0145] Historical crude oil exploration data is converted to meet the input requirements of the enhanced anchor-free detection model and normalized to produce processed exploration data. This data is then fed into the enhanced anchor-free detection model, where it undergoes forward propagation calculations and feature extraction at each layer to produce feature map data. Keypoint and size predictions are performed on the feature map data. Features from different layers are fused, and predictions that do not meet the preset requirements are removed. The resulting predictions are converted to images and integrated to produce crude oil feature image data.

[0146] The image detection-based crude oil exploration method provided in this embodiment utilizes an anchor-free detection model combined with optimization of the Giant Armadillo algorithm. This allows the model to adapt to different types of crude oil exploration image data, more accurately extracting crude oil feature image data from historical crude oil exploration data. Combined with analysis and processing of geological structure image data and surface image data, it comprehensively determines whether crude oil exists in the target area, improving the efficiency and accuracy of crude oil exploration, reducing blind exploration attempts, and lowering exploration costs. Furthermore, feature extraction and fusion of geological structure image data and surface image data yield complete feature image data containing rich information on geological structure, surface morphology, and other aspects. This provides a more comprehensive reference for geological experts and prospectors, facilitates a deeper understanding of the geological conditions in the target area, and provides a more robust basis for crude oil exploration decisions.

[0147] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0148] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A crude oil exploration method based on image detection, characterized in that: include: Collect geological structure image data and historical crude oil exploration data within the target area, and collect surface image data at preset time intervals; Constructing an anchor-free frame detection model, and optimizing the anchor-free frame detection model using a giant armadillo algorithm to obtain an enhanced anchor-free frame detection model, inputting the historical crude oil exploration data, and outputting crude oil feature image data; Analyzing the geological structure image data using a feature analysis method to obtain key feature image data, extracting surface image features from the surface image data using a multi-scale geometric analysis feature method, and combining the key feature image data with the surface image features using a feature fusion method to obtain complete feature image data; The step of extracting the surface image features includes: Converting the color image in the surface image data into a grayscale image and performing denoising processing to obtain surface image processed data; Using Curvelet to perform multi-scale decomposition on the surface image processing data, and performing directional decomposition at each scale to obtain Curvelet coefficients; Statistical analysis is performed on the Curvelet coefficients of each scale and direction to extract relevant features. Curvelet coefficients smaller than a preset threshold are removed, and geometric structure features are extracted based on the distribution and changes of the Curvelet coefficients. The surface image features are obtained by fusing the relevant features extracted at different scales and the geometric structure features extracted in different directions; The complete characteristic image data is compared with the crude oil characteristic image data to determine whether crude oil exists in the target area.

2. The crude oil exploration method based on image detection according to claim 1, characterized in that: The steps of constructing the anchor-free frame detection model include: Extracting basic features of images in the crude oil exploration data, and learning the basic features through residual blocks to build a network structure; At the output layer of the network structure, a heat map is obtained through a convolution operation, and a center point of the network structure is predicted. At the position of the center point, the width and height of the network structure are predicted through a convolution layer, and the key points of the target area are determined based on the center point, the width, and the height; Focal-Loss is used to train the heat map to calculate the center point loss; Use L1-Loss to calculate the difference between the actual value and the preset value to get the size loss; The weighted sum of the center point loss and the size loss is taken as the total loss function.

3. The crude oil exploration method based on image detection according to claim 1, characterized in that: The steps of obtaining the enhanced anchor-free detection model include: The giant armadillo population size is set, each individual armadillo represents a set of parameters of the anchor-free box detection model, and the parameter value of each individual armadillo is randomly initialized within a preset value range; For each armadillo individual, a new digging individual is generated through digging behavior; For each armadillo individual, a new escape individual is generated by escaping from perceived threats; Calculating the fitness values ​​of the initial armadillo individual, the digging individual, and the escaping individual, and selecting the armadillo individual with the highest fitness value as the individual for the next iteration; When the preset number of iterations is reached, the parameter value represented by the armadillo individual with the highest fitness value is selected and applied to the anchor-free frame detection model to obtain the enhanced anchor-free frame detection model.

4. The crude oil exploration method based on image detection according to claim 2, characterized in that: The step of outputting the crude oil characteristic image data comprises: Converting the historical crude oil exploration data into data that meets the input requirements of the enhanced anchor-free frame detection model, and performing normalization processing to obtain exploration processing data; Inputting the exploration and processing data into the enhanced anchor-free box detection model, performing forward propagation calculation through each layer, and performing feature extraction to obtain feature map data; Predicting key points and sizes of the feature map data, fusing features from different layers, and removing predictions that do not meet preset requirements to obtain prediction results; The prediction results are converted into image form and integrated to obtain crude oil characteristic image data.

5. The crude oil exploration method based on image detection according to claim 4, characterized in that: The step of obtaining the characteristic map data includes: In the first layer of the network structure, a convolution operation is performed on the exploration and processing data to obtain a convolution result, and an activation function is used to perform nonlinear changes to obtain an output feature map; Repeating the convolution operation and the activation function operation in the middle layer of the network structure until a preset number of times is reached and high-level abstract features are extracted; At the last layer of the network structure, the output feature map and the high-level abstract features are output as the feature map data.

6. The crude oil exploration method based on image detection according to claim 1, characterized in that: The step of obtaining the key feature image data includes: Using a Gaussian filter algorithm to perform denoising on the geological structure image data, and using a histogram equalization method to enhance contrast; Performing edge detection, texture analysis, and morphological feature extraction on the processed geological structure image data to obtain multiple target geological structure features, and calculating the correlation between each target geological structure feature to obtain a correlation relationship; Screening each target geological structure according to the correlation relationship and analysis purpose to obtain key features; The key features are visualized, converted into image form, and integrated to obtain the key feature image data.

7. The crude oil exploration method based on image detection according to claim 6, characterized in that: The steps to obtain the correlation relationship include: A plurality of target geological structure features are represented as vectors, and the Pearson correlation coefficient is used to calculate the correlation between each two target geological structure feature vectors to obtain a correlation matrix, and the correlation relationship is obtained by analyzing the correlation matrix.

8. The crude oil exploration method based on image detection according to claim 1, characterized in that: The steps of combining and obtaining the complete feature image data include: Standardizing the key feature image data and the surface image features, and using a maximum pooling method to make the dimensions consistent; Obtaining preset weights of the key feature image data and the surface image features according to a principal component analysis method; Different weights are assigned to the key feature image data and the surface image features according to the preset weights, and weighted addition is performed to obtain the complete feature image data.

9. The crude oil exploration method based on image detection according to claim 8, characterized in that: The step of obtaining the preset weight includes: Combining the key feature image data and the surface image features into a feature matrix, and performing a standardization process to obtain a standard feature matrix; Calculating the covariance matrix of the standard characteristic matrix and performing eigendecomposition to obtain multiple eigenvalues ​​and corresponding eigenvectors; Arrange multiple eigenvalues ​​in descending order, select a preset number of eigenvalues ​​and corresponding eigenvectors to form a principal component matrix, and calculate the contribution weights of each key feature image data and each surface image feature in the principal component matrix to obtain the preset weights.

Images