Wireless charging system foreign matter detection method combining machine vision and auxiliary coil

Abstract

A foreign matter detection method of a wireless charging system combining machine vision and auxiliary coils, comprising the following steps: 1. Making a data set; 2. Improving a YOLOv8 detection model; 3. Collecting images before wireless charging of an electric vehicle, transmitting the images to a computer terminal, using the improved YOLOv8 detection model for foreign matter detection on the computer terminal, and returning the detection result to a charging coil controller; if foreign matter is judged to exist, charging is prohibited, and charging can continue after the foreign matter is removed; if there is no foreign matter, charging is started; 4. When the electric vehicle starts charging, a planar central symmetric auxiliary coil is used for metal foreign matter detection, and the detection result is returned to the charging coil controller; if the planar central symmetric auxiliary coil detects metal foreign matter, charging is immediately stopped; if no metal foreign matter is detected, charging continues until charging is completed. The present application can effectively eliminate the detection blind area, improve the efficiency of wireless charging of electric vehicles, and ensure the safety of wireless charging.

Description

Technical Field

[0001] This invention relates to the field of metal foreign object detection technology in wireless charging systems, and mainly to a foreign object detection method for wireless charging systems that combines machine vision and auxiliary coils. Background Technology

[0002] In a wireless charging system for electric vehicles, the primary transmitting coil is placed on the ground, while the secondary receiving coil is placed on the vehicle. There is usually a large air gap between the transmitting and receiving coils, and this area may contain foreign metal objects such as coins or aluminum cans. This not only reduces the charging efficiency of the system but also poses a great safety hazard. Therefore, accurate detection of foreign objects is a necessary means to improve the safety of wireless charging systems for electric vehicles.

[0003] Currently, the main methods for foreign object detection in wireless charging systems include auxiliary coil foreign object detection technology, system parameter foreign object detection technology, and machine learning foreign object detection technology. Auxiliary coil detection technology involves placing a detection coil above the transmitting coil. When a metallic foreign object approaches the detection area, the inductance and voltage of the detection coil change, thus detecting the foreign object. A blind-spot-free metallic foreign object detection method for electric vehicle wireless charging systems with incomplete compensation of the detection coil, published by Yang Ziyue, Xia Chenyang, Zhao Shuze, et al., improves detection sensitivity by optimizing the incomplete compensation circuit parameters of the detection coil. However, auxiliary coil detection technology requires a certain amount of space, and its detection accuracy is affected to some extent when the transmitting and receiving coils are not aligned.

[0004] Foreign object detection technology for system parameters mainly detects certain system parameters such as coil quality factor, voltage, inductance, and resonant frequency, and determines whether a metallic foreign object has entered the system based on changes in these parameters. A paper by JAFARI H et al., which detects foreign objects by measuring the resonant frequency and resonant current deviation of the primary coil, proposes a discrimination mechanism that can effectively distinguish between foreign object entry and parameter changes caused by the horizontal or vertical alignment of the primary and secondary coils. A drawback of this detection technology is that when the metallic foreign object entering the wireless charging system is small, the resulting system parameters are difficult to detect, thus affecting the system's detection accuracy.

[0005] The metal foreign object detection method based on system parameters faces the problem of low detection accuracy when detecting small objects. Therefore, the method proposed in this invention can realize foreign object detection throughout the wireless charging process and has higher detection accuracy than previous methods. Summary of the Invention

[0006] To address the aforementioned technical problem of low monitoring accuracy when dealing with small metal objects, this technical solution provides a foreign object detection method for wireless charging systems that combines machine vision and an auxiliary coil. It utilizes an improved YOLOv8 network model and a planar centrally symmetrical auxiliary coil for foreign object detection. Machine learning-based foreign object detection technology acquires images of the transmitting coil surface to achieve automated detection and identification of foreign objects. The planar centrally symmetrical auxiliary coil enables accurate detection of metallic foreign objects that significantly impact the wireless charging system, providing crucial safety assurance for electric vehicle wireless charging systems. This effectively solves the aforementioned problems.

[0007] This invention is achieved through the following technical solution: A method for foreign object detection in a wireless charging system combining machine vision and an auxiliary coil, comprising the following steps: Step 1: Create a dataset. Select a dataset of wireless charging systems for electric vehicles and divide the labeled images in the dataset into training, validation, and test sets. Step 2: Improve the YOLOv8 detection model to obtain the improved YOLOv8-PDSW network model; the specific operation method is as follows: Step 2.1: Replace the network structure of the YOLOv8 model with the PP-LCNet network; Step 2.2: In the YOLOv8n backbone network, the second-to-last and third-to-last C2f modules in the backbone network are introduced into the deformable convolutional network DCN, and the modules they belong to are named C2f-DCN; Step 2.3: Add SENet attention mechanism after the YOLOv8 upsampling stage structure and after the first C2f in the downsampling stage; Step 2.4: Introduce the Wise-IoU loss function to replace the original DF Loss and CIoU Loss functions; Step 3: Before wireless charging of the electric vehicle, images are captured by video equipment and transmitted to the computer terminal. The computer terminal uses the YOLOv8-PDSW network model to detect foreign objects and returns the detection results to the charging coil controller. If a foreign object is detected, charging is prohibited. Charging can continue after the foreign object is removed. If there is no foreign object, charging begins. Step 4: When the electric vehicle starts charging, a planar centrally symmetrical auxiliary coil is used to detect metal foreign objects and the detection results are returned to the charging coil controller. If the planar centrally symmetrical auxiliary coil detects a metal foreign object, charging is stopped immediately. If no metal foreign object is detected, charging continues until charging is completed.

[0008] Furthermore, in the dataset of the electric vehicle wireless charging system described in step one, the samples in the dataset include coins, cans, nuts, screws, etc., and the number of data for all foreign objects accounts for the same proportion in the dataset; the dataset is divided into training set, validation set and test set in a 6:2:2 ratio using random sampling, and the training set and test set are labeled using data labeling tools.

[0009] Furthermore, the operation flow of the SENet attention mechanism described in step 2.3 is as follows: S1: The input feature map is transformed into a channel-level description, and global average pooling is performed on each channel to capture global contextual information; S2: The dependency relationship between learning channels is realized through two fully connected layers, and a weight is generated for each channel; S3: Adjust the original feature map using channel weights. The feature map of each channel is multiplied by its corresponding weight, thereby weakening or strengthening certain feature channels.

[0010] Furthermore, the operation flow of the Wise-IoU loss function in step 2.4 is as follows: calculate the IoU score between the ground truth box and the predicted box to measure the degree of overlap between the two, calculate the distance between the center points of the bounding boxes of the ground truth box and the predicted box, determine the maximum distance between the two, and calculate the region between the two boxes. The role of this region is to measure the relationship between the boxes.

[0011] Furthermore, the improved YOLOv8-PDSW network model uses CBS for convolution operations, DCN for deformable convolutional modules, SPPF for spatial pyramid pooling modules, SENet for attention mechanisms, Upsample for upsampling modules, Concat for feature connection modules, and Detect for detection heads.

[0012] Furthermore, the specific steps for foreign object detection using the YOLOv8-PDSW network model on the computer terminal, as described in step three, include: Step 3.1: Preprocess the image to be detected, including adjusting the image size and performing data augmentation operations; Step 3.2: Pass the processed image to the lightweight PP-LCNet network, and extract features from the input image through convolutional and pooling layers; Step 3.3: Perform bounding box prediction. Use anchor boxes on each feature map to predict the bounding box of the foreign object. Regress and classify the anchor boxes to obtain the location and category of the foreign object. Step 3.4: Identify and label the metallic foreign object target in the image based on the location information and category of the bounding box.

[0013] Furthermore, the planar centrally symmetrical auxiliary coil described in step four is based on the principle of electromagnetic induction. The alternating magnetic field will generate a certain induced voltage inside the auxiliary coil. As can be seen from the symmetry of the magnetic field of the primary transmitting coil, when the wireless charging system is operating normally, the induced voltages of a set of symmetrical auxiliary coils A1 and B1 are the same, and the difference between the two induced voltages is 0. When a metal foreign object falls on the auxiliary coil A1 or B1, the induced voltage across the auxiliary coil A1 or auxiliary coil B1 will change, thereby generating an induced voltage difference. By comparing the changes in the induced voltage difference, the rapid detection of the metal foreign object can be achieved.

[0014] Furthermore, the auxiliary coils A1 and B1 represent a pair of coils. If the auxiliary coils A1 and B1 are symmetrically arranged, when a metal foreign object is located in the middle of the auxiliary coils A1 and B1, and the metal foreign object covers the same area of ​​the two auxiliary coils, the induced voltage difference will still be 0, forming a detection blind zone. Therefore, auxiliary coil A1 is placed in the upper left corner of the plane, and auxiliary coil B1 is placed in the lower right corner of the plane. The structure of other auxiliary coils such as auxiliary coils A2 and B2 is constructed in the same way.

[0015] Furthermore, the auxiliary coils A1 and B1 form a coil pair, with the induced voltages in opposite directions and equal magnitudes under the same structure; the corresponding output voltage is U. A1 and U B1 When there are no metallic foreign objects, |U A1 |=|U B1 When a metallic foreign object falls above the auxiliary coil A1, then |U A1 |>|U B1 |or|U A1 |<|U B1 |

[0016] Furthermore, the detection circuit of the planar centrally symmetrical auxiliary coil includes an amplification circuit, a filtering circuit, and a voltage acquisition circuit. The output terminal of the centrally symmetrical auxiliary coil is connected to the amplification circuit. The voltage value obtained by electromagnetic induction is relatively small. After being amplified by the operational amplifier circuit, the voltage is input to the filtering circuit. The transmission frequency of the transmitting coil in the planar centrally symmetrical auxiliary coil is 85kHz. The filtering circuit selects the 85kHz signal to pass and suppresses signals of other frequencies. The voltage acquisition circuit acquires the voltage of the filtered circuit to obtain the induced voltage value for auxiliary detection. Beneficial effects

[0017] The present invention proposes a foreign object detection method for a wireless charging system that combines machine vision and an auxiliary coil. Compared with existing technologies, it has the following advantages: (1) This invention combines an improved YOLOv8 network model and a planar centrally symmetrical auxiliary coil for foreign object detection. The machine learning foreign object detection technology acquires images of the transmitting coil surface to achieve automated detection and identification of foreign objects. The planar centrally symmetrical auxiliary coil can accurately detect metallic foreign objects that have a significant impact on the wireless charging system, providing important protection for the safety of the electric vehicle wireless charging system. Before the wireless charging stage of the electric vehicle, the improved YOLOv8 is used for foreign object detection to ensure charging safety before wireless charging and avoid repeated vehicle start-stop. During the charging process of the electric vehicle, the planar centrally symmetrical auxiliary coil is used for foreign object detection, which can accurately detect whether there are micro-sized metallic foreign objects affecting the wireless charging system; and can quickly provide feedback on the detection results. Combining the two detection methods can not only realize the detection during the working state of the electric vehicle wireless charging system, but also perform detection in advance before the system works, realizing the detection of foreign objects in the entire stage of electric vehicle wireless charging, ensuring charging safety and improving charging efficiency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0019] Figure 2 This is a block diagram of the improved YOLOv8 network model in this invention.

[0020] Figure 3 This is a schematic diagram of the accuracy curve of the original YOLOv8 network model.

[0021] Figure 4 This is a schematic diagram of the accuracy curve of the improved YOLOv8 network model in this invention.

[0022] Figure 5 This is a schematic diagram of the precision and recall curves of the original YOLOv8 network model.

[0023] Figure 6 This is a precision and recall curve of the improved YOLOv8 network model in this invention.

[0024] Figure 7 This is a verification diagram of the experimental detection effect of the improved YOLOv8 network model in this invention.

[0025] Figure 8 This is a schematic diagram of the planar centrally symmetrical auxiliary coil in this invention.

[0026] Figure 9 This is a schematic diagram of a symmetrical auxiliary coil at the center of a plane where a metallic foreign object falls.

[0027] Figure 10The voltage waveform is generated when a metallic foreign object, such as a one-yuan coin, falls onto the detection coil. Detailed Implementation

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. The described embodiments are merely some embodiments of the present invention, and not all embodiments. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the design concept of the present invention should fall within the protection scope of the present invention. Example

[0029] like Figure 1 As shown, a foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil includes the following steps: Step 1: Create a dataset. Select a dataset of wireless charging systems for electric vehicles and divide the labeled images in the dataset into training, validation, and test sets.

[0030] The dataset for the wireless charging system for electric vehicles includes samples such as coins, cans, nuts, and screws, with all foreign objects having an equal proportion of data. The dataset is divided into training, validation, and test sets in a 6:2:2 ratio using random sampling, and the training and test sets are labeled using data annotation tools.

[0031] The training set is used to train the improved model, applying gradient descent to the training error during training, and the training weights can be adjusted experimentally. The validation set is used to tune the model's hyperparameters and to perform an initial evaluation of the model's capabilities. The test set is used to check the model's detection accuracy and evaluate its generalization ability.

[0032] Step 2: Improve the YOLOv8 detection model to obtain the improved YOLOv8-PDSW network model; the specific operation method is as follows: Step 2.1: Replace the network structure of the YOLOv8 model with the PP-LCNet network.

[0033] PP-LCNet is a lightweight convolutional neural network based on a deep neural network mathematical core library, designed to improve computational efficiency and performance on mobile devices; it is more suitable for practical application scenarios due to its advantages such as low computational cost and high efficiency.

[0034] The Darknet-53 module of the YOLOv8 backbone network has a deep network structure, which can improve the model's ability to represent image features to a certain extent. However, due to the complexity of its modules, the model's inference time becomes longer. Therefore, this embodiment uses the PP-LCNet module in YOLOv8n to improve the backbone network in order to achieve a lightweight and fast detection method.

[0035] Step 2.2: In the YOLOv8n backbone network, the second-to-last and third-to-last C2f modules in the backbone network are introduced into the deformable convolutional network DCN, and the modules they belong to are named C2f-DCN.

[0036] Deformable convolutions can adaptively adjust the shape and size of the receptive field based on the irregular shape of the object, further improving the robustness of CNNs. Traditional convolution operations can only extract features from regular rectangular receptive fields, but in actual wireless charging processes, foreign objects entering the system are often irregular, so deformable convolutions are needed to identify irregularly shaped objects. During sampling, deformable convolutions more closely match the shape of the foreign object and are more robust. Furthermore, in electric vehicle wireless charging systems, small objects can easily fall in; if traditional convolutions are still used, some small objects may be difficult to detect. Therefore, as... Figure 2 As shown, the second-to-last and third-to-last C2f modules in the backbone network are replaced with deformable convolutional modules to improve the detection performance of small target foreign objects.

[0037] Step 2.3: Add the SENet attention mechanism after the YOLOv8 upsampling stage structure and after the first C2f in the downsampling stage.

[0038] The SENet network can automatically learn the importance of features in different channels. SENet's attention mechanism enables the model to better focus on foreign object features, improving the accuracy of foreign object detection. SENet represents a channel attention strategy that can automatically learn the importance of features in different channels; simply put, it can enhance important features and suppress unimportant features. Figure 2 As shown, SENet attention mechanism is added after the upsampling stage structure and after the first C2f in the downsampling stage to improve the detection effect of foreign objects.

[0039] The operation flow of the SENet attention mechanism is as follows: S1: The input feature map is transformed into a channel-level description, and global average pooling is performed on each channel to capture global contextual information; S2: The dependency relationship between learning channels is realized through two fully connected layers, and a weight is generated for each channel; S3: Adjust the original feature map using channel weights. The feature map of each channel is multiplied by its corresponding weight, thereby weakening or strengthening certain feature channels.

[0040] Step 2.4: Introduce the Wise-IoU loss function to replace the original DF Loss and CIoU Loss functions.

[0041] YOLOv8n uses DF Loss and CIoU Loss to calculate bounding box regression losses. CIoU uses a monotonic focusing mechanism but does not consider the balance between easy and hard samples. Furthermore, traditional loss functions only consider the intersection-union ratio (IoU) of the predicted and ground truth boxes during detection, neglecting classification information. Therefore, to address the bias caused by the traditional IoU loss function, this embodiment introduces the Wise-IoU loss function.

[0042] The operation process of the Wise-IoU loss function is as follows: calculate the IoU score between the ground truth box and the predicted box to measure the degree of overlap between the two, calculate the distance between the center points of the bounding boxes of the ground truth box and the predicted box, determine the maximum distance between the two, and calculate the region between the two boxes. The role of this region is to measure the relationship between the boxes.

[0043] like Figure 2 As shown, the improved YOLOv8-PDSW network model uses CBS for convolution operations, DCN for deformable convolutional modules, SPPF for spatial pyramid pooling modules, SENet for attention mechanisms, Upsample for upsampling, Concat for feature connection modules, and Detect for the detection head.

[0044] To verify the superiority of the improved YOLOv8-PDSW network model, the inventors conducted simulation verification and analysis, the specific process of which is as follows: (a) Creating a dataset: Since there is currently no publicly available and comprehensive dataset of the required type on the Internet, a dataset was created using online resources and manual photography. This dataset includes four categories, selecting coins, aluminum cans, screws, and nuts as metallic foreign objects, totaling 1500 images.

[0045] The dataset was randomly selected and divided into a training set (1050 images), a validation set (300 images), and a test set (150 images) in a 7:2:1 ratio. The training set was used to train the improved model, applying gradient descent to the training error and setting the trainable weight parameters. The validation set was used to adjust the model's hyperparameters and perform a preliminary evaluation of its capabilities. The test set was used to check the model's detection accuracy and evaluate its generalization ability.

[0046] (II) Setting of evaluation indicators

[0047] This embodiment uses precision, number of parameters (M), mean average precision (mAP), F1-score, and floating-point operations (FLOPs) to evaluate the improved model.

[0048] Accuracy refers to the ratio of correctly detected samples to the total number of detected samples. Parameter count refers to the number of parameters in the model. Mean accuracy (mAP) represents the mean, and mAP@0.5 refers to the average accuracy across all categories in an object detection task when the Intersection over Union (IoU) reaches 0.5. Floating-point computation count is the number of floating-point operations per second, and it is one of the standards for evaluating the computational load of a network model.

[0049] (III) Simulation Analysis

[0050] This simulation experiment was conducted on a 64-bit Windows 11 operating system, with an RTX 3090 GPU with 24GB of VRAM, an Intel Core i7-10700k CPU, 96GB of RAM, Python 3.9.13 as the programming language, CUDA version 12.2.146 as the CUDA version, a batch size of 10, and 150 iterations. Training was performed using the deep learning framework PyTorch 1.11.0.

[0051] 3.1 Comparison and Evaluation of Different Models and Their Results To verify the differences and superiority of the improved algorithm in this embodiment compared with other object detection algorithms, precision, recall, mAP@0.5, and floating-point operations were used as evaluation metrics. The improved model YOL0v8-PDSW in this embodiment was compared with the YOL0v5, YOL0v7, YOLOv8n, and YOLOv8s algorithms under the same conditions. The precision curves of YOL0v8n and YOL0v8-PDSW are shown below. Figure 3 and Figure 4 As shown in Table 1, the simulation results for each evaluation index are presented.

[0052] according to Figure 3 and Figure 4 In comparison, the accuracy of the original model before the improvement was 92.6%, while the accuracy of the improved model after the improvement was 95.7%, representing a 3.1% improvement in accuracy compared to the original model.

[0053] Table 1

[0054] According to the simulation results in Table 1, the unmodified YOLOv8n algorithm integrates the advantages of previous single-target detection algorithms, surpassing the previous YOLOv5 and YOLOv7 algorithms in accuracy, mAP@0.5%, and F1-Score. Furthermore, the YOLOv8n algorithm has fewer parameters and fewer floating-point computations than the YOLOv5 and YOLOv7 algorithms. While the YOLOv8s algorithm outperforms the YOLOv8n algorithm in mAP@0.5%, its accuracy is lower, and it significantly exceeds YOLOv8n in terms of parameter count and floating-point computations.

[0055] 3.2 Comparative Analysis of Results with Different Improved Modules To verify the effectiveness of each improved module of the algorithm in this embodiment, the YOL0v8n model was used as the baseline model. Precision, recall, mAP@0.5%, and floating-point computation were used as evaluation indicators. The results were compared by permutation and combination of different improved modules. The results are shown in Table 2.

[0056]

[0057] To better verify the impact of the improved modules and the combination of improved modules on the performance improvement of the original YOLOv8n algorithm, this experiment designed a comparative analysis, and the simulation results are shown in Table 2.

[0058] Where √ indicates that the module has been added, and × indicates that it has not been added. After improving the original YOLOv8n backbone network by using the PP-LCNet lightweight network, the accuracy decreased by 3.4% and mAP@0.5% decreased by 1.8% compared to the original algorithm. The number of model parameters and floating-point computations were significantly reduced. Adding DCN deformable convolution and SE attention mechanism to the original YOLOv8n algorithm model significantly improved the accuracy and slightly improved mAP@0.5%, but this increased the complexity of the model, resulting in a larger number of parameters and floating-point computations.

[0059] Adding the SE attention mechanism and the Wise-IoU loss function to the original YOLOv8n algorithm improves accuracy to a certain extent and effectively enhances the model's bounding box regression ability. However, mAP@0.5% decreases slightly, and the number of parameters and floating-point calculations increases. In summary, the improved YOLOv8-PDSW model studied in this embodiment achieves a 3.1% improvement in accuracy and a 0.6% improvement in mAP@0.5 compared to the original YOLOv8n algorithm model. The number of parameters and floating-point calculations decreases by 26% and 13%, respectively, demonstrating the effectiveness of the proposed improved algorithm. The precision and recall curves of YOLOv8n and YOLOv8-PDSW are shown below. Figure 5 and Figure 6 As shown.

[0060] Depend on Figure 5 and Figure 6 The comparison shows that the improved algorithm improves the overall mAP@0.5% by 0.6%, remains basically stable in the coin and nut categories, and shows improvement in the screw and aluminum can categories.

[0061] 3.3 Verification of Experimental Detection Results like Figure 7 As shown, the original YOLOv8n algorithm model does not perform particularly well when dealing with foreign objects of varying sizes and complex backgrounds. To verify that the improved YOLOv8-PDSW algorithm model still achieves better detection results when facing these problems, this paper conducts comparative experiments. Confidence represents the probability that the detection system judges the detected object as a foreign object; a higher confidence level indicates a greater probability. Comparing Experiment 1 and Experiment 2, in Experiment 1, the detection confidence of the YOLOv8n-PDSW model increased from 0.5 to 0.7 compared to the original YOLOv8 model, indicating that the improved model is more likely to judge the object as a foreign object. In Experiment 2, the detection confidence of the YOLOv8n-PDSW model is higher than that of the original YOLOv8n model. Therefore, the effectiveness of the improved model can be concluded.

[0062] in conclusion This embodiment employs an improved YOLOv8 model. Based on the original YOLOv8 model, the improved model replaces the backbone network with a lightweight PP-LCNet network to reduce the number of parameters and floating-point computations. It introduces deformable convolutional networks (DCN) to enhance the ability to identify irregularly shaped foreign objects, adds a SE attention mechanism to improve feature fusion capabilities, and uses Wise-IoU as the loss function to improve the network's bounding box regression ability. Compared to the original model, this improved model achieves a 3.1% improvement in detection accuracy, a 26% reduction in model parameters, and a 13% reduction in floating-point computations. Therefore, the method proposed in this embodiment can meet the requirements for foreign object detection in electric vehicle wireless charging systems and has practical application value.

[0063] Step 3: Before wireless charging of the electric vehicle, images are captured by video equipment and transmitted to the computer terminal. The computer terminal uses the YOLOv8-PDSW network model to detect foreign objects and returns the detection results to the charging coil controller. If a foreign object is detected, charging is prohibited. Charging can continue after the foreign object is removed. If there is no foreign object, charging begins.

[0064] The specific steps for foreign object detection using the YOLOv8-PDSW network model on a computer terminal include: Step 3.1: Preprocess the image to be detected, including adjusting the image size and performing data augmentation operations; Step 3.2: Pass the processed image to the lightweight PP-LCNet network, and extract features from the input image through convolutional and pooling layers; Step 3.3: Perform bounding box prediction. Use anchor boxes on each feature map to predict the bounding box of the foreign object. Regress and classify the anchor boxes to obtain the location and category of the foreign object. Step 3.4: Identify and label the metallic foreign object target in the image based on the location information and category of the bounding box.

[0065] Step 4: When the electric vehicle starts charging, a planar centrally symmetrical auxiliary coil is used to detect metallic foreign objects, and the detection result is returned to the charging coil controller. The detection circuit of the planar centrally symmetrical auxiliary coil includes an amplification circuit, a filtering circuit, and a voltage acquisition circuit. The output terminal of the centrally symmetrical auxiliary coil is connected to the amplification circuit. The voltage value obtained by electromagnetic induction is relatively small. After being amplified by the operational amplifier circuit, the voltage is input to the filtering circuit. The transmission frequency of the transmitting coil in the planar centrally symmetrical auxiliary coil is 85kHz. The filtering circuit selects the 85kHz signal to pass and suppresses other frequency signals. The voltage acquisition circuit acquires the voltage of the filtered circuit to obtain the induced voltage value of the auxiliary detection.

[0066] The detection principle of a planar centrally symmetrical detection coil is based on electromagnetic induction. When a metallic foreign object enters the wireless charging system, it causes a change in the magnetic field of the system, thereby generating an induced electromotive force. When the metallic foreign object falls above the auxiliary coil, the equivalent inductance and equivalent resistance of the centrally symmetrical auxiliary coil change accordingly, resulting in a change in the output induced voltage of the coil. Different types and sizes of metallic foreign objects have different equivalent inductances and equivalent resistances, affecting the degree of change in the parameters of the centrally symmetrical auxiliary coil to varying degrees. The more drastic the change in the parameters of the centrally symmetrical auxiliary coil, the easier it is to detect.

[0067] The planar centrally symmetrical auxiliary coil is based on the principle of electromagnetic induction. The alternating magnetic field will generate a certain induced voltage inside the auxiliary coil. Due to the symmetry of the magnetic field of the primary transmitting coil, when the wireless charging system is operating normally, the induced voltages of a set of symmetrical auxiliary coils A1 and B1 are the same, and the difference between the two induced voltages is 0. When a metal foreign object falls on the auxiliary coil A1 or B1, the induced voltage across the auxiliary coil A1 or auxiliary coil B1 will change, thereby generating an induced voltage difference. By comparing the changes in the induced voltage difference, rapid detection of metal foreign objects can be achieved.

[0068] like Figure 8As shown, the planar centrally symmetrical auxiliary coil consists of 5 groups of coils, namely A1, B1, A2, B2, A3, B3, A4, B4, A5, and B5. All coils are in close contact with the surface of the transmitting coil. The planar centrally symmetrical auxiliary coil, the transmitting coil, and the receiving coil all use rectangular coils.

[0069] Auxiliary coils A1 and B1 represent a pair of coils. If auxiliary coils A1 and B1 are symmetrically arranged, when a metal foreign object is located in the middle of auxiliary coils A1 and B1, and the metal foreign object covers the same area of ​​the two auxiliary coils, the induced voltage difference will still be 0, forming a detection blind zone. Therefore, auxiliary coil A1 is placed in the upper left corner of the plane, and auxiliary coil B1 is placed in the lower right corner of the plane. The structure of other auxiliary coils such as auxiliary coils A2 and B2 is constructed in the same way.

[0070] Auxiliary coils A1 and B1 form a coil pair. The induced voltages in the two coils are in opposite directions, and under the same structure, the induced voltages are equal in magnitude; the corresponding output voltage is U. A1 and U B1 When there are no metallic foreign objects, |U A1 |=|U B1 When a metallic foreign object falls above the auxiliary coil A1, then |U A1 |>|U B1 |or|U A1 |<|U B1 |

[0071] When a metallic foreign object enters the wireless charging system and lands precisely between auxiliary coil A1 and auxiliary coil B5, such as Figure 9 As shown, the black object represents a metallic foreign object. Auxiliary coil A1 in the upper left corner and auxiliary coil B1 in the lower right corner form one set of auxiliary coils. Auxiliary coil B5 in the upper right corner and auxiliary coil A5 in the lower left corner form another set of auxiliary coils. The induced voltage of auxiliary coil A1 will change, while the induced voltage of auxiliary coil B1 remains constant. The induced voltage of auxiliary coil B5 will change, while the induced voltage of auxiliary coil A5 remains constant. At this point, there is an induced voltage difference, and the differential circuit output is not zero, indicating the presence of a metallic foreign object in the system. The calculation formula is:

[0072] in, This represents the induced voltage difference formed by auxiliary coil A1 and auxiliary coil B1. This represents the induced voltage of the auxiliary coil A1. This represents the induced voltage of the auxiliary coil B1. This represents the induced voltage difference formed by auxiliary coil A5 and auxiliary coil B5. This represents the induced voltage of the auxiliary coil A5. This represents the induced voltage of the auxiliary coil B5.

[0073] At t=0.01s, a metal foreign object, a one-yuan coin, enters the coupling mechanism of the wireless charging system. When the metal foreign object, a one-yuan coin, lands above the auxiliary coil A1, Figure 10 The diagram shows the voltage waveform of auxiliary coil A1. Due to certain errors in the circuit design, the detection system has an output voltage of approximately 0.1V. The threshold voltage for the detection system used to determine the presence of metallic foreign objects is set to 0.1V. Figure 10 It can be seen that there are no metal foreign objects in the coupling mechanism before 0.01s. When a 1 yuan coin foreign object is present, the output voltage of the detection system becomes 1.1V, which is higher than the set voltage threshold. Therefore, it can be determined that there are metal foreign objects in the coupling mechanism of the wireless charging system.

[0074] If the planar center-symmetric auxiliary coil detects a metallic foreign object, charging will stop immediately; if no metallic foreign object is detected, charging will continue until charging is complete.

[0075] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered by the present invention.

Claims

1. A foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil, characterized in that, Including the following steps: Step 1: Create a dataset. Select a dataset of wireless charging systems for electric vehicles and divide the labeled images in the dataset into training, validation, and test sets. Step 2: Improve the YOLOv8 detection model to obtain the improved YOLOv8-PDSW network model; the specific operation method is as follows: Step 2.1: Replace the network structure of the YOLOv8 model with the PP-LCNet network; Step 2.2: In the YOLOv8n backbone network, the second-to-last and third-to-last C2f modules in the backbone network are introduced into the deformable convolutional network DCN, and the modules they belong to are named C2f-DCN; Step 2.3: Add the SENet attention mechanism after the YOLOv8 upsampling stage structure and after the first C2f in the downsampling stage; the operation flow of the SENet attention mechanism is as follows: S2.31: The input feature map is transformed into a channel-level description, and global average pooling is performed on each channel to capture global contextual information; S2.32: The dependency relationship between learning channels is realized through two fully connected layers, and a weight is generated for each channel; S2.33: Adjust the original feature map using channel weights, multiplying the feature map of each channel by its corresponding weight; Step 2.4: Introduce the Wise-IoU loss function to replace the original DF Loss and CIoU Loss functions; the operation flow of the Wise-IoU loss function is as follows: Calculate the IoU score between the ground truth box and the predicted box to measure their overlap. Calculate the distance between the center points of the bounding boxes of the ground truth box and the predicted box to determine the maximum distance between them. Calculate the region between the two boxes, which is used to measure the relationship between the boxes. Step 3: Before wireless charging of the electric vehicle, images are captured by video equipment and transmitted to the computer terminal. The computer terminal uses the YOLOv8-PDSW network model to detect foreign objects and returns the detection results to the charging coil controller. If a foreign object is detected, charging is prohibited. Charging can continue after the foreign object is removed. If there is no foreign object, charging begins. Step 4: When the electric vehicle starts charging, a planar centrally symmetrical auxiliary coil is used to detect metallic foreign objects, and the detection results are returned to the charging coil controller; If the planar center-symmetric auxiliary coil detects a metallic foreign object, charging will stop immediately; if no metallic foreign object is detected, charging will continue until charging is complete.

2. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 1, characterized in that: In the dataset of the electric vehicle wireless charging system described in step one, the samples in the dataset include coins, cans, nuts, and screws, and the number of data for each foreign object is equal in the dataset. The dataset is divided into training set, validation set and test set in a 6:2:2 ratio using random sampling, and the training set and test set are labeled using data labeling tools.

3. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 1, characterized in that: The improved YOLOv8-PDSW network model uses CBS for convolution operations, DCN for deformable convolutional modules, SPPF for spatial pyramid pooling modules, SENet for attention mechanisms, Upsample for upsampling, Concat for feature connection modules, and Detect for detection heads.

4. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 1, characterized in that: Step three describes using the YOLOv8-PDSW network model on a computer terminal for foreign object detection. The specific steps include: Step 3.1: Preprocess the image to be detected, including adjusting the image size and performing data augmentation operations; Step 3.2: Pass the processed image to the lightweight PP-LCNet network, and extract features from the input image through convolutional and pooling layers; Step 3.3: Perform bounding box prediction. Use anchor boxes on each feature map to predict the bounding box of the foreign object. Regress and classify the anchor boxes to obtain the location and category of the foreign object. Step 3.4: Identify and label the metallic foreign object target in the image based on the location information and category of the bounding box.

5. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 1, characterized in that: The planar centrally symmetrical auxiliary coil described in step four is based on the principle of electromagnetic induction. The alternating magnetic field will generate a certain induced voltage inside the auxiliary coil. As can be seen from the symmetry of the magnetic field of the primary side transmitting coil, when the wireless charging system is operating normally, the induced voltages of a set of symmetrical auxiliary coils A1 and B1 are the same, and the difference between the two induced voltages is 0. When a metallic foreign object falls onto auxiliary coil A1 or B1, the induced voltage across auxiliary coil A1 or auxiliary coil B1 will change, thereby generating an induced voltage difference. By comparing the changes in the induced voltage difference, the metallic foreign object can be detected quickly.

6. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 5, characterized in that: The auxiliary coils A1 and B1 represent a pair of coils. Auxiliary coil A1 is placed in the upper left corner of the plane, and auxiliary coil B1 is placed in the lower right corner of the plane. The structure formed by auxiliary coils A2 and B2 and other auxiliary coils is similar.

7. A foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil, as described in claim 5 or 6, characterized in that: The auxiliary coils A1 and B1 form a coil pair, with the induced voltages in opposite directions and equal magnitudes under the same structure; the corresponding output voltage is U. A1 and U B1 When there are no metallic foreign objects, |U A1 |=|U B1 When a metallic foreign object falls above the auxiliary coil A1, then |U A1 |>|U B1 |or|U A1 |<|U B1 | 8. The foreign object detection method for a wireless charging system combining machine vision and an auxiliary coil according to claim 5, characterized in that: The detection circuit of the planar centrally symmetrical auxiliary coil includes an amplification circuit, a filtering circuit, and a voltage acquisition circuit. The output terminal of the centrally symmetrical auxiliary coil is connected to the amplification circuit. The voltage value obtained by electromagnetic induction is relatively small. After being amplified by the operational amplifier circuit, the voltage is input to the filtering circuit. The transmission frequency of the transmitting coil in the planar centrally symmetrical auxiliary coil is 85KHz. The filtering circuit selects the 85KHz signal to pass and suppresses signals of other frequencies. The voltage acquisition circuit acquires the voltage of the filtered circuit to obtain the induced voltage value for auxiliary detection.

Images