Yolov4-based gpr image root point recognition and wave velocity evaluation method

Abstract

The application relates to a GPR image root point recognition and wave velocity evaluation method based on Yolov4. The method comprises the following steps: firstly, a root system scene dataset is constructed; then, a GPR image root point recognition and wave velocity evaluation model is constructed based on a Yolov4 target detection algorithm, wherein the model comprises a core feature extraction part, a feature processing and fusion part and a target prediction part; then, the root system scene dataset is input into the GPR image root point recognition and wave velocity evaluation model for training, and key point coordinates and wave velocity prediction values are output; finally, hyperbolic curve fitting is carried out based on the key point coordinates, and root point positions and propagation speeds are obtained. The method solves the problems of difficult acquisition of measured samples, slow generation of simulated samples and single sample richness, improves the automation level of root system recognition and wave velocity evaluation, and reduces the time cost of simulated image data generation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of environmental science and technology, in particular to a GPR image root point recognition and wave velocity evaluation method based on Yolov4. BACKGROUND

[0002] Plant roots, as an indispensable part of the ecosystem, can efficiently absorb water and nutrients in the soil, and play a crucial role in maintaining plant life activities and promoting ecosystem functions. Nowadays, ground penetrating radar (GPR) has been successfully applied to root research as a geophysical method. It determines the distribution of underground medium by transmitting electromagnetic waves through a transmitting antenna and receiving underground echo signals through a receiving antenna. Compared with traditional destructive methods such as root excavation and soil profile method, GPR can repeatedly and non-destructively detect roots. When GPR is used for ground detection, a detection image B-scan can be obtained, which is a two-dimensional image of GPR dragging distance and electromagnetic wave propagation time. Root points often appear as hyperbolic reflection shapes in the B-scan image. In ground penetrating radar root research, root point recognition and wave velocity extraction are crucial. The identification of hyperbolic curves can be used to determine the distribution of root points in the B-scan image. The determination of electromagnetic wave velocity can obtain root depth information on the basis of automatic root point recognition, and can obtain soil moisture information through the Topp formula.

[0003] In traditional methods, Hough transformation, least squares method and other methods are usually used to identify and locate hyperbolic curves on the B-scan image, or Canny filter is used to detect hyperbolic curve contour edge points, and the vertex information of the hyperbolic curve is obtained by fitting calculation using the analytical fitting function of the hyperbolic curve, or random Hough transformation (RHT) is used to identify root points and obtain the position, depth and velocity information of the root points. This method first needs to give a reasonable velocity range, and then identifies the hyperbolic curve set on this basis, and finally performs manual screening. However, the above traditional methods have high precision but high time cost, which is not conducive to quickly processing rich data collected in the field.

[0004] Therefore, in the related art, there is an urgent need for a method that can improve the efficiency of ground penetrating radar image root identification and reduce the time cost. SUMMARY

[0005] Therefore, in the related art, there is an urgent need for a method that can improve the efficiency of ground penetrating radar image root identification and reduce the time cost.

[0006] In a first aspect, the application provides a GPR image root point recognition and wave velocity evaluation method based on Yolov4. The method comprises:

[0007] constructing a root scene dataset;

[0008] constructing a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, the model including a core feature extraction part, a feature processing and fusion part, and a target prediction part;

[0009] inputting the root scene dataset into the GPR image root point recognition and wave speed evaluation model for training, and outputting key point coordinates and wave speed prediction values;

[0010] performing hyperbolic curve fitting based on the key point coordinates to obtain root point positions and propagation speeds.

[0011] Optionally, in an embodiment of the present application, the construction of the root scene dataset comprises:

[0012] obtaining a single-root scene feature map, and designing a Merge strategy to fuse to generate a multi-root scene image;

[0013] performing style transfer on the multi-root scene image based on a Multi-CycleGAN network to obtain the root scene dataset.

[0014] Optionally, in an embodiment of the present application, the design of the Merge strategy to fuse to generate a multi-root scene image comprises:

[0015] adjusting the contrast and translation properties of the feature map to simulate forward simulation images under different root water content and horizontal position changes.

[0016] Optionally, in an embodiment of the present application, the performing of style transfer on the multi-root scene image based on the Multi-CycleGAN network further comprises:

[0017] based on a root-free reflection image in a measured environment as a background, performing style transfer supplement using the Merge strategy.

[0018] Optionally, in an embodiment of the present application, the target prediction part is constructed based on a prediction head adding key point and wave speed prediction regression head.

[0019] Optionally, in an embodiment of the present application, the hyperbolic curve fitting based on the key point coordinates comprises:

[0020] determining the root point position based on the vertex key point, performing parameter reduction processing on a hyperbolic reflection shape formula, and performing least square secondary fitting on other coordinates except the vertex key point to calculate the propagation speed.

[0021] Optionally, in an embodiment of the present application, the method further comprises:

[0022] Evaluate the hyperbolic curve, the recognition effect of key points and the prediction effect of wave speed of the GPR image root point recognition and wave speed evaluation model.

[0023] In a second aspect, the application further provides a GPR image root point recognition and wave speed evaluation device based on Yolov4. The device comprises:

[0024] A data set construction module for constructing a root system scene data set;

[0025] A GPR image root point recognition and wave speed evaluation model construction module for constructing a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, wherein the model comprises a core feature extraction part, a feature processing and fusion part and a target prediction part;

[0026] A model training module for inputting the root system scene data set into the GPR image root point recognition and wave speed evaluation model for training, and outputting key point coordinates and wave speed prediction values;

[0027] A root point position and propagation speed determination module for performing root point hyperbolic fitting based on the key point coordinates to obtain the root point position and the propagation speed.

[0028] In a third aspect, the application further provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the steps of the method described in each of the above embodiments.

[0029] In a fourth aspect, the application further provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the method described in each of the above embodiments.

[0030] The above-mentioned GPR image root point recognition and wave velocity evaluation method based on Yolov4 first constructs a root system scene dataset; then, a GPR image root point recognition and wave velocity evaluation model is constructed based on a Yolov4 target detection algorithm, the model including a core feature extraction part, a feature processing and fusion part and a target prediction part; then, the root system scene dataset is input into the GPR image root point recognition and wave velocity evaluation model for training, and key point coordinates and wave velocity prediction values are output; finally, based on the key point coordinates, a root point hyperbolic curve fitting is performed to obtain the root point position and propagation speed. That is, the root system scene dataset construction framework based on the Merge and Multi-CycleGAN strategies is designed to generate the data required for model training, solving the problems of difficult acquisition of measured samples, slow generation of simulated samples and single sample richness, and a Yolov4-HPV deep learning model is constructed to realize root point recognition and wave velocity evaluation in GPR detection images. A key point module is developed, and the five key points recognized by Yolov4-HPV are used to fit the root point hyperbolic curve as a supplement to the wave velocity prediction result, improving the automation level of root system recognition and wave velocity evaluation and reducing the time cost of simulated image data generation. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 An application environment diagram of the GPR image root point recognition and wave velocity evaluation method based on Yolov4 in one embodiment;

[0032] Figure 2 A flowchart of the GPR image root point recognition and wave velocity evaluation method based on Yolov4 in one embodiment;

[0033] Figure 3 A schematic diagram of constructing a root system scene dataset in one embodiment;

[0034] Figure 4 A schematic diagram of a single root feature map in one embodiment;

[0035] Figure 5 A schematic diagram of root distribution features in one embodiment;

[0036] Figure 6 A double-root style migration result diagram in one embodiment;

[0037] Figure 7 A structural diagram of the GPR image root point recognition and wave velocity evaluation model in one embodiment;

[0038] Figure 8 A schematic diagram of GPR detecting underground roots in one embodiment;

[0039] Figure 9 A detection result of the model on the test set in one embodiment;

[0040] Figure 10 A structural block diagram of a GPR image root point recognition and wave velocity evaluation device based on Yolov4 in an embodiment;

[0041] Figure 11 An internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0042] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0043] The GPR image root point recognition and wave velocity evaluation method based on Yolov4 provided by the embodiments of the present application can be applied in the application environment as shown in Figure 1 . Among them, the terminal communicates with the server through the network. The data storage system can store the data required to be processed by the server. The data storage system can be integrated on the server, or placed on the cloud or other network servers. Among them, the terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server can be realized by an independent server or a server cluster composed of multiple servers.

[0044] In an embodiment, as shown in Figure 2 , a GPR image root point recognition and wave velocity evaluation method based on Yolov4 is provided. Taking the server in Figure 1 as an example for illustration, the method comprises the following steps:

[0045] S201: Constructing a root system scene dataset.

[0046] In the embodiments of the present application, first, a GPR root system reflection acquisition pattern is obtained, and a root system scene dataset is constructed.

[0047] Specifically, in an embodiment of the present application, the construction of the root system scene dataset comprises:

[0048] S301: Obtaining a single-root scene feature map, and designing a Merge strategy to fuse and generate a multi-root scene image.

[0049] S303: Performing style transfer on the multi-root scene image based on a Multi-CycleGAN network to obtain a root system scene dataset.

[0050] In an embodiment of the present application, as Figure 3 (a) is a schematic diagram for constructing the root scene dataset. First, the propagation of electromagnetic waves in the ground is simulated by means of the gprMax forward simulation software to generate a series of different single root scene feature maps. For the case of multiple root distribution, a Merge strategy is designed to effectively fuse multiple single root scene feature maps together to quickly generate a large number of complex multi-root scene images. This strategy greatly speeds up the generation of samples and labels and effectively reduces the time cost of forward simulation and label annotation. Specifically, to learn the root point reflection characteristics under different soil relative permittivity and different depths, eight soil relative permittivity gradients are set: 3.5, 4, 4.5, 5, 5.5, 6, 6.5, and 7, and five depth gradients: 0.1, 0.2, 0.3, 0.4, and 0.5 m. The corresponding soil volume water content range is 3% to 13% calculated by the Topp formula. The scene profile size is 4.0 m x 1 m, the time window is 20 ns, the root diameter is 1 cm, the root relative permittivity is 12.01, the Ricker of 900 MHz is used as the excitation source, the distance between the transmitting and receiving antennas is 0.14 m, and the moving step is 0.01 m. A total of 40 single root feature maps are generated by gprMax forward simulation, and each map has a width representing 4 m and a height representing 20 ns of electromagnetic wave propagation time. As shown in Figure 4 , the shape and opening of the hyperbolic reflection signal of the root point differ under different soil permittivity and depth.

[0051] Secondly, the Multi-CycleGAN network is used for style transfer of the multi-root scene image to enhance the generalization ability of the model and obtain more realistic GPR root reflection acquisition images in the field to constitute the root scene dataset. CycleGAN is a kind of generative adversarial network that can learn the conversion of source domain images to target domain images. This process only relies on two different domain datasets without strict corresponding relationship between the data. The unique advantage of CycleGAN is that it can preserve the key root shape and opening feature information in the simulation data while only transferring the style elements of the image, ensuring that the subsequent model can still effectively identify and analyze these root features in the field data. To increase the diversity of the output, different training cycle model weight files are selected to constitute the Multi-CycleGAN for style transfer of the image. As shown in Figure 3 (c), CycleGAN is composed of a forward network (source domain X to target domain Y) and a reverse network (target domain Y to source domain X). The generator G maps the image in the source domain X to the target domain space to obtain the output ​​The loss function thereof includes two parts of a generative adversarial loss and a cycle loss. Specifically, taking multiple scene image data as a source domain X and shrub GPR measured images collected in Hunshandake sandy land in Inner Mongolia as a target domain Y, model weight files of 66th, 67th, 76th, 77th, 91st, 95th and 100th epochs are selected according to the size of the loss function. As shown in Figure 6 Fig. 6, the result images under different style transfer are displayed taking double roots as an example. It can be seen that the background heterogeneity of the image after style transfer is stronger, and can better represent the noise caused by the field soil. At the same time, the hyperbolic shape and the opening size signal are effectively preserved. The background ripples, hyperbolic brightness and heterogeneity of each result image are inconsistent, which enriches the patterns of root samples and helps to improve the robustness of the model.

[0052] In an embodiment of the present application, the design Merge strategy fusion generates multiple root scene images, which includes:

[0053] The contrast and translation properties of the feature map are adjusted to simulate forward simulation images under different root water content and horizontal position changes.

[0054] In an embodiment of the present application, considering the diversity of field root distribution morphology, four representative root distribution characteristics under natural conditions are summarized from measured images, as shown in Figure 5 (a), which correspond to four designed root distributions, including single root distribution, double root distribution, triangular three root distribution and ladder-shaped three root distribution, as shown in Figure 5 (b). For the case of multiple root distribution, as shown in Figure 3 (b), the contrast between the hyperbolic curve and the background is adjusted by histogram equalization to simulate the change of root water content; and the image matrix position is adjusted by image translation to simulate the change of root horizontal position. The translation formula is as follows.

[0055]

[0056] Wherein, T(I, λ) is the left or right translation of the image matrix I by λ pixels, H is the image matrix width, and cat is the matrix splicing function.

[0057] The adjusted image is fused with other single root feature images of the same depth to realize the automatic generation of multiple root distribution images.

[0058] In an embodiment of the present application, the style transfer of the multiple root scene images based on the Multi-CycleGAN network further includes:

[0059] Based on the root-free reflection image in the measured environment as the background, the Merge strategy is used for style transfer supplement.

[0060] In an embodiment of the present application, the measured environment without root system reflection image is selected as the background, and the designed Merge strategy is used for style transfer supplement to make it more consistent with the real GPR root system reflection acquisition pattern in the field.

[0061] S203: Construct a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, the model including a core feature extraction part, a feature processing and fusion part, and a target prediction part.

[0062] In an embodiment of the present application, a Yolov4 target detection framework is used as the main architecture, a key point method module is developed, and a GPR image root point recognition and wave speed evaluation model Yolov4-HPV (Yolov4-Hyperbolic Position and Velocity) is constructed. Specifically, the model includes three components: a core feature extraction part BackBone, a feature processing and fusion part Neck, and a target prediction part Prediction. As shown in Figure 7 As shown in the input end, the Yolov4-HPV receives a 480x480, 3-channel format GPR detection image as input data. The core feature extraction part BackBone uses an efficient CSP (Convolutional Block Shortcuts) structure, as shown in Figure 7 (b), which divides the feature map of the base layer into two independent paths, and then fuses the information through a cross-stage merging mechanism, effectively reducing the memory consumption required for calculation, while retaining rich feature information. The Neck part integrates two modules: SPP (Spatial Pyramid Pooling) and PANet (Path Aggregation Network), which jointly act on the depth processing and fusion of the feature map. As shown in Figure 7(c), the SPP module significantly expands the receptive field of the feature map by applying multi-scale processing to the upper layer feature map with multiple size (13x13, 9x9, 5x5, 1x1) pooling kernels, effectively capturing the context features at different scales. The PANet module further strengthens the feature fusion strategy by constructing a bottom-up feature transmission path, incorporating more shallow detail information such as edge shape, while combining bottom-up feature aggregation to achieve deep fusion of high and low layer features, significantly improving the network's detection ability for small size and complex shape targets. The Prediction part as the prediction end of Yolov4-HPV uses multi-scale feature map parallel processing to adapt to the detection needs of different size root hyperbolic curves. The output dimension is set to 17 dimensions, including the coordinates of the target rectangular frame, class labels, confidence, five key point coordinates and wave speed prediction values. This design enables the model to further evaluate the corresponding wave speed information based on accurate identification of root hyperbolic curves in GPR images.

[0063] Specifically, in an embodiment of the present application, the target prediction part is constructed based on adding a key point and wave speed prediction regression head.

[0064] In an embodiment of the present application, as shown in Figure 7 (a), a key point and wave speed prediction regression head is added to the prediction head of Yolov4, the original prediction output of Yolov4 is 6 dimensions, including target rectangular frame coordinates, class labels and confidence, and after adding the key point and wave speed prediction regression head, the output dimension of the Yolov4-HPV model becomes 17 dimensions, which increases 5 key point coordinates and wave speed prediction values on the basis of the original.

[0065] S205: inputting the root system scene data set into the GPR image root point recognition and wave speed evaluation model for training, and outputting key point coordinates and wave speed prediction values.

[0066] In the embodiments of the present application, the constructed root system scene data set is input into the GPR image root point recognition and wave speed evaluation model for training. Specifically, a total of 5250 images with a size of 480x480 are generated, including 406 single-root images, 1260 double-root images, 1536 triangular three-root images, and 2048 ladder three-root images, a total of 13678 root targets. 70% of the training data is used to train the model, 10% of the validation data is used to adjust the model parameters, and 20% of the test data is used to evaluate the model. The YOLOv4-HPV model is trained using the PyTorch open source neural network framework, the batch size is set to 6 during training, the initial learning rate is set to 0.01, and a total of 200 epochs are trained. After training, the prediction results including key point coordinates and wave speed prediction values are output.

[0067] S207: Perform hyperbolic fitting based on the key point coordinates to obtain the root point position and propagation speed.

[0068] In the embodiments of the present application, the five key points identified by the Yolov4-HPV model are used to perform hyperbolic fitting on the root point to obtain the root point position and propagation speed, which are used as a supplement to the wave speed evaluation result.

[0069] Specifically, in one embodiment of the present application, the hyperbolic fitting based on the key point coordinates includes:

[0070] Based on the vertex key point to determine the root point position, the hyperbolic reflection shape formula is reduced in parameters, and the least square secondary fitting is performed on the coordinates other than the vertex key point to calculate the propagation speed.

[0071] In one embodiment of the present application, as shown in the formula (1), the measurement line direction is perpendicular to the root growth direction, and when the GPR is dragged along the ground measurement line, due to the characteristics of the conical diffusion of the electromagnetic wave emitted by the transmitting antenna, the root point will receive the reflection signal from the root system in front and behind. When the antenna is located directly above the root point, the propagation path of the transmitted signal is the shortest, so a typical hyperbolic reflection shape is formed on the B-scan image, and the mathematical expression of the hyperbolic reflection shape can be described by the following formula according to the radar propagation principle. Figure 8

[0072]

[0073] Wherein, x0, t0 are the vertex of the hyperbola, that is, the coordinates of the root point in the B-scan image, and v is the propagation speed of the electromagnetic wave between the root point and the ground.

[0074] There are a total of 3 unknown parameters x0, t0 and v in the formula, the third key point in the output of the Yolov4-HPV model is the vertex key point, which can locate the hyperbolic root point. The vertex key point is used as the true root point position, and the hyperbolic reflection shape formula is reduced in parameters, which can be described by the following formula in the image, and then the least square secondary fitting is performed on the remaining 4 key points to calculate the propagation speed v.

[0075]

[0076] Wherein, X and Y are pixel coordinates in the picture, X0 and Y0 are pixel coordinates of the root point predicted by the model, which are known values, and k and m are distances and times represented by each pixel.

[0077] In one embodiment of the present application, the method further includes:

[0078] ​The hyperbolic curve, the identification effect of the key point, and the prediction effect of the wave velocity of the GPR image root point identification and wave velocity evaluation model are evaluated.

[0079] In an embodiment of the present application, the Yolov4-HPV model is evaluated by three types of indexes. The first type of P, R, and AP IOU＝0.5 P is the accuracy, R is the recall rate, and AP IOU＝0.5 is the average precision when IOU=0.5, and AP is the area between the precision-recall curve and the coordinate axis, and IOU is the intersection over union of the predicted bounding box and the true bounding box; the second type of AP OKS＝0.5 is used to evaluate the identification effect of the key point, OKS is similar to IOU, and is used to calculate the similarity of the predicted key point and the target key point. The third type of RMSE is the root mean square error, which is used to evaluate the effect of the predicted wave velocity value. The deviation degree of the predicted value and the true value is measured by calculating the square root of the sum of squares of the difference between the sample predicted value and the true value. The specific formula is as follows:

[0080]

[0081] Among them, TP is a true sample, that is, the model predicts true, and the sample is also true; FP is a false sample, that is, the model predicts true, and the sample is false; FN is a missed sample, that is, the model predicts false, and the sample is actually true.

[0082]

[0083] Among them, v i is the visibility of the i th key point; δ(v i >0) is a judgment function, which is 1 when the condition is met, otherwise 0; d i represents the Euclidean distance between the i th predicted key point and the target key point; S is the square root of the area occupied by the root point hyperbola; σ i is the normalization factor of the i th key point.

[0084]

[0085] Among them, N is the sample number of the corresponding data set; V i is the true wave velocity value corresponding to the root point, is the wave velocity value evaluated by Yolov4-HPV or the wave velocity value calculated by the key point method.

[0086] In an embodiment of the present application, the Yolov4-HPV model has excellent performance in a total of 1050 test set images, and the average detection time of each image is about 5 milliseconds, wherein P, R, AP IOU＝0.5 and the key point evaluation index AP OKS ＝0.5All reached above 0.95. As shown in Figure 9 The recognition results of Yolov4-HPV in the test set are shown, including hyperbolic prediction box, 5 key points and wave velocity, where the lm value is the wave velocity value calculated by the key point method, and the blue curve is the hyperbola fitted by the key point method. From the Figure 8 The results show that no matter the single root, double roots and different combinations of three roots, the Yolov4-HPV model can well locate the hyperbola and the 5 key points on the hyperbola, and the hyperbola fitted by the key point method is very consistent with the actual hyperbola, indicating that the 5 key points positioning has accuracy, and the calculated wave velocity has great credibility.

[0087] In an embodiment of the present application, as shown in the following table, the wave velocity values predicted by Yolov4-HPV and the key point method under each soil dielectric constant are shown, and the true value of the wave velocity is obtained by converting the soil dielectric constant. Among the number of roots with a soil dielectric constant of 6, Yolov4-HPV and the key point method perform best, where the RMSE of the wave velocity predicted by Yolov4-HPV is 0.0036 m / ns, and the average relative error is 2.47%, and the RMSE of the key point method is 0.0039 m / ns, and the average relative error is 2.55%. Among the number of roots with a soil dielectric constant of 7, Yolov4-HPV and the key point method have the largest error, but the average relative error is less than 10%, where the RMSE of the wave velocity predicted by Yolov4-HPV is 0.0121 m / ns, and the average relative error is 9.88%, and the RMSE of the key point method is 0.0058 m / ns, and the average relative error is 4.17%. Overall, the average relative errors of Yolov4-HPV and the key point method are less than 5%, which are 4.31% and 3.06% respectively. Overall, the performance of Yolov4-HPV and the key point method on the test set shows that Yolov4-HPV can not only accurately locate the hyperbola but also accurately evaluate the wave velocity value, and the 5 key points identified by Yolov4-HPV can also well calculate the wave velocity value.

[0088] Table Wave velocity evaluation results of different soil relative dielectric constants

[0089]

[0090] In the above-mentioned GPR image root point recognition and wave velocity evaluation method based on Yolov4, first, a root system scene dataset is constructed; then, a GPR image root point recognition and wave velocity evaluation model is constructed based on a Yolov4 target detection algorithm, the model including a core feature extraction part, a feature processing and fusion part, and a target prediction part; then, the root system scene dataset is input into the GPR image root point recognition and wave velocity evaluation model for training, and key point coordinates and wave velocity prediction values are output; finally, root point hyperbolic curve fitting is performed based on the key point coordinates to obtain root point positions and propagation velocities. That is, the root system scene dataset construction framework based on Merge and Multi-CycleGAN strategies is designed to generate data required for model training, solving the problems of difficult acquisition of measured samples, slow generation of simulated samples, and single sample richness, a Yolov4-HPV deep learning model is constructed, root point recognition and wave velocity evaluation in GPR detection images are realized, a key point module is developed, and the five key points recognized by Yolov4-HPV are used to fit the root point hyperbolic curve as a supplement to the wave velocity prediction result, improving the automation level of root system recognition and wave velocity evaluation and reducing the time cost of simulated image data generation.

[0091] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in sequence according to the direction of the arrow. Unless otherwise explicitly stated herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0092] Based on the same inventive concept, the embodiments of the present application also provide a Yolov4-based GPR image root point recognition and wave velocity evaluation device for implementing the above-mentioned Yolov4-based GPR image root point recognition and wave velocity evaluation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more Yolov4-based GPR image root point recognition and wave velocity evaluation device embodiments provided below can be referred to the limitations of the Yolov4-based GPR image root point recognition and wave velocity evaluation method described above, which will not be repeated here.

[0093] In one embodiment, as Figure 10As shown, a Yolov4-based GPR image root point recognition and wave speed evaluation device 1000 is provided, comprising a dataset construction module 1001, a GPR image root point recognition and wave speed evaluation model construction module 1003, a model training module 1005, and a root point position and propagation speed determination module 1007, wherein:

[0094] The dataset construction module 1001 is configured to construct a root system scene dataset.

[0095] The GPR image root point recognition and wave speed evaluation model construction module 1003 is configured to construct a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, wherein the model comprises a core feature extraction part, a feature processing and fusion part, and a target prediction part.

[0096] The model training module 1005 is configured to input the root system scene dataset into the GPR image root point recognition and wave speed evaluation model for training, and output key point coordinates and wave speed prediction values.

[0097] The root point position and propagation speed determination module 1007 is configured to perform root point hyperbolic curve fitting based on the key point coordinates to obtain root point position and propagation speed.

[0098] In an embodiment of the present application, the dataset construction module is further configured to:

[0099] Obtain a single-root scene feature map, design a Merge strategy to fuse and generate a multi-root scene image;

[0100] Perform style transfer on the multi-root scene image based on a Multi-CycleGAN network to obtain a root system scene dataset.

[0101] In an embodiment of the present application, the dataset construction module is further configured to:

[0102] Adjust the contrast and translation properties of the feature map to simulate forward simulation images under different root system water content and horizontal position changes.

[0103] In an embodiment of the present application, the dataset construction module is further configured to:

[0104] Based on the measured environment without root system reflection images as the background, the Merge strategy is used for style transfer supplement.

[0105] In an embodiment of the present application, the target prediction part is constructed based on a prediction head adding key point and wave speed prediction regression head.

[0106] In an embodiment of the present application, the root point position and propagation speed determination module is further configured to:

[0107] Based on the vertex key point, the root point position is determined, the hyperbolic reflection shape formula is reduced, and the least square secondary fitting is performed on the coordinates other than the vertex key point to calculate the propagation velocity.

[0108] The Yolov4-based GPR image root point identification and wave velocity evaluation device further includes a model evaluation module, which is further used for:

[0109] Evaluating the identification effect of hyperbolic curves and key points and the prediction effect of wave velocity of the GPR image root point identification and wave velocity evaluation model.

[0110] The above-mentioned various modules in the Yolov4-based GPR image root point identification and wave velocity evaluation device can be all or part realized by software, hardware and combinations thereof. The above-mentioned various modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above-mentioned various modules.

[0111] In one embodiment, a computer device is provided, which can be a terminal, and the internal structure diagram thereof can be as shown in Figure 11 The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement a Yolov4-based GPR image root point identification and wave velocity evaluation method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0112] Those skilled in the art can understand that Figure 11 the structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0113] In an embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0114] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0115] In an embodiment, a computer program product is provided, comprising a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0116] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric memory (Ferroelectric Random Access Memory, FRAM), phase change memory (Phase Change Memory, PCM), graphene memory, etc. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0118] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0119] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A Yolov4-based GPR image root point recognition and wave velocity evaluation method, characterized in that, The method comprises: constructing a root system scene dataset; constructing a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, the model comprising a core feature extraction part, a feature processing and fusion part, and a target prediction part, the target prediction part being constructed based on a prediction head adding key points and a wave speed prediction regression head; inputting the root system scene dataset into the GPR image root point recognition and wave speed evaluation model for training, and outputting key point coordinates and wave speed prediction values; performing hyperbolic curve fitting based on the key point coordinates to obtain root point positions and propagation speeds; the construction of the root system scene dataset comprises: obtaining a single-root scene feature map, and designing a Merge strategy to fuse and generate a multi-root scene image; performing style transfer on the multi-root scene image based on a Multi-CycleGAN network to obtain a root system scene dataset; the hyperbolic curve fitting based on the key point coordinates comprises: determining root point positions based on vertex key points, performing parameter reduction processing on a hyperbolic curve reflection shape formula, and performing least squares secondary fitting on coordinates other than the vertex key points to calculate propagation speeds.

2. The Yolov4-based GPR image root point recognition and wave velocity evaluation method according to claim 1, characterized in that, the design of the Merge strategy to fuse and generate the multi-root scene image comprises: adjusting the contrast and translation properties of the feature map to simulate forward simulation images under different root system water contents and horizontal position changes.

3. The Yolov4-based GPR image root point recognition and wave velocity evaluation method according to claim 1, characterized in that, the style transfer on the multi-root scene image based on the Multi-CycleGAN network further comprises: using the Merge strategy to perform style transfer and supplement based on a root-free reflection image in a measured environment as a background.

4. The Yolov4-based GPR image root point recognition and wave velocity evaluation method according to claim 1, characterized in that, The method further comprises: evaluating the identification effect of hyperbolic curves, key points, and the prediction effect of wave speeds of the GPR image root point recognition and wave speed evaluation model.

5. A Yolov4-based GPR image root point recognition and wave velocity evaluation device, characterized in that, The device comprises: a dataset construction module for constructing a root system scene dataset; a GPR image root point recognition and wave speed evaluation model construction module for constructing a GPR image root point recognition and wave speed evaluation model based on a Yolov4 target detection algorithm, the model comprising a core feature extraction part, a feature processing and fusion part, and a target prediction part, the target prediction part being constructed based on a prediction head adding key points and a wave speed prediction regression head; a model training module for inputting the root system scene dataset into the GPR image root point recognition and wave speed evaluation model for training, and outputting key point coordinates and wave speed prediction values; a root point position and propagation speed determination module for performing hyperbolic curve fitting based on the key point coordinates to obtain root point positions and propagation speeds; the construction of the root system scene dataset comprises: obtaining a single-root scene feature map, and designing a Merge strategy to fuse and generate a multi-root scene image; performing style transfer on the multi-root scene image based on a Multi-CycleGAN network to obtain a root system scene dataset; the hyperbolic curve fitting based on the key point coordinates comprises: determining root point positions based on vertex key points, performing parameter reduction processing on a hyperbolic curve reflection shape formula, and performing least squares secondary fitting on coordinates other than the vertex key points to calculate propagation speeds. 6.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-5 when the computer program is executed by the processor. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 4.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 4.

Images