Unmanned aerial vehicle focus point selection method based on YOLO V5

Through YOLO V5 combining attention mechanism and FPN module's drone source point selection method, the problems of poor excitation effect and data loss caused by improper point selection in traditional methods are solved, and intelligent placement point selection is realized, improving the source excitation effect and exploration data quality promotes the efficiency and accuracy of geophysical exploration.

CN119027844AInactive Publication Date: 2024-11-26CHINA UNIV OF GEOSCIENCES (BEIJING)
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411507509.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-28
Publication Date
2024-11-26
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional drone seismic source point selection methods rely on manual survey and map data, making it difficult to consider surface conditions, resulting in poor excitation effects and reduced seismic data quality, and cannot cover areas that are difficult to reach by manpower.

Method used

Using the YOLO V5-based drone source point selection method, combined with the target detection algorithm and geophysical exploration needs, through the attention mechanism and FPN module, intelligent placement point selection is realized, accurately identify topography and geological structure information, and optimize placement Click Select.

Benefits of technology

It improves the accuracy and reliability of point selection, improves the excitation effect of drone sources and the quality of exploration data, optimizes the efficiency and accuracy of geophysical exploration, and reduces costs and risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027844A_ABST
    Figure CN119027844A_ABST
Patent Text Reader

Abstract

The invention discloses an unmanned aerial vehicle focus point selection method based on YOLO V5, and the method is characterized in that the method comprises the following steps: S1, collecting target region image data, and carrying out the preprocessing of the data, and obtaining a data set; s2, constructing a YOLO V5 detection model based on an attention mechanism; s3, inputting the data set into an attention mechanism-based YOLO V5 detection model for model training, evaluation and testing; and S4, verifying a throwing condition: judging whether a target location in the target area is used as an alternative throwing area of the unmanned aerial vehicle seismic source or not. According to the unmanned aerial vehicle seismic source point selection method based on YOLO V5, the computer vision technology and seismic exploration requirements are combined, an attention mechanism is introduced, a small target detection probe is added, and mosaic enhancement is adopted in the data loading process, so that the recognition accuracy and robustness of an unmanned aerial vehicle on a seismic source point in a complex environment can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of geophysical exploration, and in particular to a UAV earthquake source point selection method based on YOLO V5. Background Art

[0002] In the field of geophysical exploration, choosing the right drop point is crucial to the stimulation effect of UAV seismic sources. Traditional methods mainly rely on manual surveys and map data. However, this method has some limitations, especially when the influence of surface conditions on the stimulation effect is not considered during the point selection process, which may lead to poor stimulation effect and reduced seismic data quality.

[0003] At the same time, since the landing point of the drone earthquake source may involve areas that are difficult to reach by manpower, the feasibility of traditional methods is limited. Summary of the invention

[0004] To solve the above problems, the present invention provides a UAV source point selection method based on YOLO V5. By combining the target detection algorithm and the needs of geophysical exploration, the limitations of traditional methods can be overcome, and intelligent release point selection can be realized. It can accurately identify information such as topography, geological structure, etc., improve the accuracy and reliability of point selection, and help avoid problems such as poor excitation effect or data missing caused by improper point selection, thereby improving the quality and availability of exploration data, thereby improving the excitation effect of the UAV source, optimizing the efficiency and accuracy of geophysical exploration, providing more reliable data support for geophysical exploration, and promoting the further development and application of exploration work.

[0005] To achieve the above object, the present invention provides a UAV earthquake source selection method based on YOLO V5, comprising the following steps: S1, collecting image data of the target area and preprocessing it to obtain a data set; S2. Build a YOLO V5 detection model based on the attention mechanism; S3, input the data set into the YOLO V5 detection model based on the attention mechanism for model training, evaluation and testing; S4. Verify the throwing conditions: determine whether the target location in the target area is an alternative throwing area for the UAV seismic source.

[0006] Preferably, the preprocessing steps described in step S1 are: image cropping and scaling, data enhancement, annotation, annotation verification and data segmentation; Among them, image cropping and scaling is used to crop and scale the acquired target area image to adapt to the input size of the model; Data augmentation, used to increase the diversity of image data through mosaic enhancement techniques; Annotation, used to annotate and classify image data according to image information; Annotation verification, used to verify the annotation information to obtain the data set; Data splitting is used to randomly divide the dataset into training, validation and test sets.

[0007] Preferably, in step S2, the attention module is a SENet module or a CBAM module and a NLNN module; When the attention module is a SENet module, the SENet module is placed after the downsampling layer of the Backbone module; When the attention modules are the CBAM module and the NLNN module, the NLNN module is placed after the C3 module of the Backbone module, and the CBAM module is placed after the C3 module of the neck module.

[0008] Preferably, the depth of the NLNN module is set to 1, the initial number of channels is set to 64, and the initial attention is set to 8; The scaling of the CABM module is set to 1, the depth is set to 1, and the ratio of channel attention and spatial attention is set to 0.67.

[0009] Preferably, step S3 specifically includes the following steps: S31, transmitting the training set to the YOLO V5 detection model based on the attention mechanism for training, and saving each training model; S32, input the validation set to the training model after every set epoch, and save the best model according to the evaluation conditions; S33, input the test set into the best model to obtain the test result.

[0010] Preferably, the evaluation criteria described in step S32 include accuracy, recall, F1 score and mAP; The accuracy calculation formula is as follows: ; In the formula, Indicates accuracy; Represents a positive sample; Represents negative samples; The recall calculation formula is as follows: ; In the formula, represents the recall rate; Represents a negative positive sample; The F1 score calculation formula is as follows: ; In the formula, Indicates accuracy.

[0011] Preferably, the throwing condition described in step S4 is for: ; In the formula, Indicates wires, The value of is 0 or 1, where 1 is present and 0 is absent; It means river, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates the road, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates house, The value of is 0 or 1, where 1 is present and 0 is absent; For vehicles, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates pedestrians, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates animals, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates a steep slope. The value of is 0 or 1, where 1 is present and 0 is absent; Represents and symbol; like =1, it means that throwing is impossible and an alarm message is issued; if =0, it means it is an alternative throwing area.

[0012] The present invention has the following beneficial effects: High accuracy and robustness: By introducing the attention mechanism and FPN module, as well as using technologies such as mosaic data enhancement, the model's recognition accuracy and robustness for small targets can be improved. The attention mechanism enables the model to better focus on important features, thereby improving the detection capability of small targets such as earthquake sources. Adapt to complex environments: Through data enhancement and model optimization, the model can adapt to complex and changing environments, including different lighting conditions, background noise, and changes in target scale and shape. At the same time, by adopting mosaic enhancement technology, the model's robustness to scale, position, and background changes is effectively increased, making it more reliable in real scenes. Efficient and real-time: The YOLO V5 model with efficient real-time detection capabilities, combined with the attention mechanism and FPN module, can achieve faster reasoning speed while maintaining high recognition accuracy, which is suitable for real-time monitoring and point selection tasks of drones; Superior overall performance: By comprehensively considering indicators such as accuracy, recall, F1 score and mAP for model training and evaluation, the model is ensured to have excellent performance in all performance indicators, thereby improving the efficiency and reliability of seismic exploration tasks; Automatic site selection: Combined with drone technology, it can realize intelligent site selection of potential earthquake source points in seismic exploration, reducing the cost and risk of manual exploration and improving the efficiency and safety of exploration; In summary, the UAV seismic source intelligent site selection method based on YOLO V5 has many technical advantages and is expected to play an important role in fields such as seismic exploration, improve exploration efficiency, reduce costs, and provide technical support for UAV seismic source intelligent site selection.

[0013] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of a method for selecting a UAV earthquake source based on YOLO V5 of the present invention; Figure 2 The YOLO V5 framework diagram of the UAV earthquake source selection method based on YOLO V5 of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention are further described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions.

[0016] It should be noted that the terms "include", "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or devices.

[0017] Like reference numerals and letters denote similar items in the following drawings, and thus, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0018] As an efficient target detection algorithm, YOLO V5 can quickly and accurately identify multiple factors such as topography, geological conditions, etc., so as to automatically identify suitable deployment points. The introduction of the above-mentioned intelligent point selection work will help improve the accuracy and efficiency of deployment point selection, optimize the geophysical exploration process, improve data quality, and reduce costs.

[0019] In terms of surface conditions, surface conditions with good excitation effects usually include flat terrain, stable geology, and low-noise environment, which are conducive to reducing the impact of terrain changes and underground medium inhomogeneity on seismic wave propagation, thereby improving the excitation effect. On the contrary, uneven terrain, complex geological structure, or high-noise environment will reduce the excitation effect and lead to a decline in the quality of seismic data.

[0020] Therefore, YOLO V5 is used to perform intelligent point selection, which can optimize the selection of drone source placement points.

[0021] Based on the above analysis, the present invention is designed as follows: Figure 1 and Figure 2 As shown, a method for selecting a seismic source of a UAV based on YOLO V5 includes the following steps: S1, collecting image data of the target area and preprocessing it to obtain a data set; Considering that most of the landing points are relatively small, the resolution of the image should be as high as possible to ensure the visibility of small targets. The preprocessing steps described in step S1 are: image cropping and scaling, data enhancement, annotation, annotation verification and data segmentation; among them, image cropping and scaling are used to crop and scale the collected target area image to adapt to the input size of the model while maintaining the possibility of recognizing small targets; data enhancement is used to increase the diversity of image data and improve the generalization ability of the model through mosaic enhancement technology. The mosaic enhancement technology is to increase the robustness of the model to scale, position and background changes by splicing four training pictures into one; annotation is used to annotate and classify image data according to image information. The annotation points of this embodiment include power lines, rivers, roads, houses, vehicles, pedestrians, animals, steep slopes, and suitable excitation points; annotation verification is used to verify the annotation information to obtain a data set; data segmentation is used to randomly divide the data set into a training set, a validation set and a test set.

[0022] S2. Build a YOLO V5 detection model based on the attention mechanism; In step S2, the attention module is a SENet module (SENet enhances the representation ability of the model by recalibrating the feature response of the channel) or a CBAM module (CBAM introduces attention in the spatial dimension, allowing the model to pay more attention to important spatial areas) and a NLNN module (the NLNN module can capture global context information, which helps to improve the model's perception of small targets); when the attention module is a SENet module, the SENet module is placed after the downsampling layer of the Backbone module to enhance the model's adaptability to features of different scales; when the attention module is a CBAM module and a NLNN module, the NLNN module is placed after the C3 module of the Backbone module, and the CBAM module is placed after the C3 module of the neck module.

[0023] In this embodiment, since the CBAM module includes attention in two dimensions, channel and space, it can be considered to be placed in a part where feature fusion is more intensive, such as the FPN structure, to optimize the ability to focus on small target features.

[0024] The depth of the NLNN module is set to 1, the initial number of channels is set to 64, and the initial attention is set to 8; the scaling ratio of the CABM module is set to 1, the depth is set to 1, and the ratio of channel attention and spatial attention is set to 0.67.

[0025] In this embodiment, Anchor Bonxes and DetectionHead are adjusted according to the size and aspect ratio of the small target, which specifically includes the following steps: First, design the experiment: Set up the initial configuration: select different attention module combinations (such as SENet+CBAM, SENet+NLNN, CBAM+NLNN), and record the evaluation metrics: accuracy, recall, F1 score, and mAP value.

[0026] Then conduct the following experiment: Step 1: Train the initial model: Use the default configuration for preliminary training and evaluate the performance.

[0027] Step 2: Module combination experiment: train and evaluate different attention module combinations separately.

[0028] Step 3. Result analysis: Select the optimal combination and adjust the Anchor Boxes.

[0029] Finally, adjust the Anchor Boxes and Detection Head: Step 1: Initial setup: Set the initial Anchor Boxes ([10x13, 16x30, 33x23]) according to the actual size of the small target in the target area.

[0030] Step 2: Iterative optimization: After each round of training, the Anchor Boxes are adjusted to ([9x14, 17x21, 24x30]) according to the detection effect of the validation set.

[0031] S3, input the data set into the YOLO V5 detection model based on the attention mechanism for model training, evaluation and testing; Step S3 specifically includes the following steps: S31, transmitting the training set to the YOLO V5 detection model based on the attention mechanism for training, and saving each training model; S32. Input the validation set to the training model after every set epoch, and save the best model according to the evaluation conditions; in this embodiment, the validation is performed every 10 epochs.

[0032] The evaluation criteria described in step S32 include accuracy, recall, F1 score and mAP; Among them, accuracy (Precision): the proportion of samples that are actually positive examples among the samples predicted by the model as positive examples. It measures the proportion of positive samples among all samples predicted by the model as positive examples. The accuracy calculation formula is as follows: ; In the formula, Indicates accuracy; Represents a positive sample; Represents negative samples; Recall refers to the proportion of samples that the model can correctly predict as positive examples. It measures how many of the real positive examples are correctly predicted by the model. The formula for calculating recall is as follows: ; In the formula, represents the recall rate; Represents a negative positive sample; F1 score: It takes into account both accuracy and recall. It is the harmonic mean of accuracy and recall. The higher the F1 score, the better the overall performance of the model. The formula for calculating the F1 score is as follows: ; In the formula, Indicates accuracy.

[0033] mAP (Mean Average Precision): It is a commonly used indicator in target detection tasks. It represents the average value of the AP of each category in the detection results of different categories. AP (Average Precision) means that the average precision of each category is the area under the Precision-Recall curve. mAP is the average AP value of all categories.

[0034] In this embodiment, in order to ensure the accuracy and comprehensiveness of the detection results, the mAP is set to be greater than 0.7. S33, input the test set into the best model to obtain the test results. The batch size is selected as 16, 32 or 64. The IoU threshold is selected as 0.6 to determine the degree of match between the predicted box and the true box. The initial learning rate is selected as 0.0001 as the starting value.

[0035] S4. Verify the throwing conditions: determine whether the target location in the target area is an alternative throwing area for the UAV seismic source.

[0036] Throwing conditions described in step S4 for: ; In the formula, Indicates wires, The value of is 0 or 1, where 1 is present and 0 is absent; It means river, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates the road, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates house, The value of is 0 or 1, where 1 is present and 0 is absent; For vehicles, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates pedestrians, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates animals, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates a steep slope. The value of is 0 or 1, where 1 is present and 0 is absent; Represents and symbol; like =1, it means that throwing is impossible and an alarm message is issued; if =0, it means it is an alternative throwing area.

[0037] Therefore, the present invention adopts the above-mentioned UAV source point selection method based on YOLO V5, and by combining the target detection algorithm and the geophysical exploration needs, it can overcome the limitations of the traditional method and realize the intelligent selection of release points. It can accurately identify the terrain, geological structure and other information, improve the accuracy and reliability of point selection, and help avoid problems such as poor excitation effect or data missing caused by improper point selection, thereby improving the quality and availability of exploration data, thereby improving the excitation effect of the UAV source, optimizing the efficiency and accuracy of geophysical exploration, providing more reliable data support for geophysical exploration, and promoting the further development and application of exploration work.

[0038] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for selecting the seismic source of a drone based on YOLO V5. Features: The following steps are involved: S1, collecting image data of the target area and preprocessing it to obtain a data set; S2. Build a YOLO V5 detection model based on the attention mechanism; S3, input the data set into the YOLO V5 detection model based on the attention mechanism for model training, evaluation and testing; S4. Verify the throwing conditions: determine whether the target location in the target area is an alternative throwing area for the UAV seismic source.

2. A UAV earthquake source selection method based on YOLO V5 according to claim 1, Features: The preprocessing steps described in step S1 are: image cropping and scaling, data enhancement, labeling, label verification and data segmentation; Among them, image cropping and scaling is used to crop and scale the acquired target area image to adapt to the input size of the model; Data augmentation, used to increase the diversity of image data through mosaic enhancement techniques; Annotation, used to annotate and classify image data according to image information; Annotation verification, used to verify the annotation information to obtain the data set; Data splitting is used to randomly divide the dataset into training, validation and test sets.

3. A UAV earthquake source selection method based on YOLO V5 according to claim 2, Features: In step S2, the attention module is a SENet module or a CBAM module and a NLNN module; When the attention module is a SENet module, the SENet module is placed after the downsampling layer of the Backbone module; When the attention modules are the CBAM module and the NLNN module, the NLNN module is placed after the C3 module of the Backbone module, and the CBAM module is placed after the C3 module of the neck module.

4. The UAV earthquake source selection method based on YOLO V5 according to claim 3, Features: The depth of the NLNN module is set to 1, the initial number of channels is set to 64, and the initial attention is set to 8; The scaling of the CABM module is set to 1, the depth is set to 1, and the ratio of channel attention and spatial attention is set to 0.

67.

5. The UAV earthquake source selection method based on YOLO V5 according to claim 4, Features: Step S3 specifically includes the following steps: S31, transmitting the training set to the YOLO V5 detection model based on the attention mechanism for training, and saving each training model; S32, input the validation set to the training model after every set epoch, and save the best model according to the evaluation conditions; S33, input the test set into the best model to obtain the test result.

6. The UAV earthquake source selection method based on YOLO V5 according to claim 5, Features: The evaluation criteria described in step S32 include accuracy, recall, F1 score and mAP; The accuracy calculation formula is as follows: ; In the formula, Indicates accuracy; Represents a positive sample; Represents negative samples; The recall calculation formula is as follows: ; In the formula, represents the recall rate; Represents a negative positive sample; The F1 score calculation formula is as follows: ; In the formula, Indicates accuracy.

7. The UAV earthquake source selection method based on YOLO V5 according to claim 6, Features: Throwing conditions described in step S4 for: ; In the formula, Indicates wires, The value of is 0 or 1, where 1 is present and 0 is absent; It means river, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates the road, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates house, The value of is 0 or 1, where 1 is present and 0 is absent; For vehicles, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates pedestrians, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates animals, The value of is 0 or 1, where 1 is present and 0 is absent; Indicates a steep slope. The value of is 0 or 1, where 1 is present and 0 is absent; Represents and symbol; like =1, it means that throwing is impossible and an alarm message is issued; if =0, it means it is an alternative throwing area.

Citation Information

Patent Citations

  • Multipoint geostatistics pre-stack inversion method based on updated probability ratio constant theory

    CN111273348A

  • Improved unmanned aerial vehicle aerial image dense and small target identification method, system and device based on YOLOv5s and medium

    CN117456389A

  • Oil-gas exploration system and method based on remote sensing technology

    CN117993735A

  • Multi-point geostatistical prestack inversion method based on renewal probability ratio constant theory

    US20210263176A1