Terrain classification method and system based on plate capacitance tomography sensor

Abstract

The invention discloses a terrain classification method and system based on a plate capacitance tomography sensor, and belongs to the technical field of robot application. The method comprises the following steps: firstly, designing a panel ECT sensor array comprising eight electrodes, and building a mobile robot body sensing data acquisition system, so as to acquire mutual capacitance data and inertial data of terrains; and secondly, performing feature extraction, fusion and classification on the capacitance data and the inertial data by building a multi-layer perceptron classification model, after the multi-layer perceptron model is trained, moving the robot to a known terrain, and completing the classification of the terrain through the multi-layer perceptron. According to the method, two kinds of modal information of inertia and capacitance are utilized, through combination of interactive vibration information of the robot and the terrains and changes of dielectric constants in the terrains, the terrains made of various different materials can be automatically recognized, and the method has high universality and practical significance.

Description

Technical Field

[0001] The present invention relates to the field of robot application technology, and in particular to a terrain classification method and system based on a flat-plate capacitance tomography sensor. Background Art

[0002] With the rapid development of robotics, the ability of robots to autonomously navigate in unknown environments has become a research hotspot. In complex environments, robots need to accurately identify and classify different terrain types in order to select appropriate paths or adjust their movement patterns, thereby improving their maneuverability and reducing bumps during movement. Robotic systems are typically equipped with a variety of sensors, primarily external perception sensors and proprioception sensors. External perception sensors (such as RGB cameras, LiDAR, etc.) can capture visual information about the terrain and reflect the differences in appearance between different terrains. However, these sensors are easily affected by lighting conditions and the atmospheric environment. For example, in strong or low light conditions, visual data may contain a large amount of noise, resulting in reduced terrain classification accuracy. In addition, terrain classification methods that rely solely on visual sensors are fragile in outdoor environments and are difficult to meet the needs of practical applications.

[0003] In order to overcome the limitations of external perception sensors, researchers have begun to explore the use of proprioception sensors for terrain classification. Proprioception sensors (such as force-torque sensors, inertial measurement units, tactile sensors, etc.) identify terrain features by measuring internal parameters (such as force, vibration, acceleration, etc.) when the robot interacts with the terrain. For example, some studies install accelerometers or tactile sensors on the robot to collect tactile data of the robot's interaction with the terrain and encode it into contextual representations to achieve terrain classification. Other studies have designed specific pressing and friction actions, combined with force-torque sensors to measure foot forces, to further improve the accuracy of terrain classification. However, the performance of proprioception sensors is highly dependent on the quality of the interaction between the robot and the terrain, and different movement patterns may have a significant impact on the perception results. In complex environments, it is difficult for the robot to always maintain a stable motion state, which may lead to a decrease in the reliability of proprioception data.

[0004] In summary, existing technologies still face many challenges in terrain classification. Therefore, developing a terrain classification method that is insensitive to environmental changes and motion patterns and has high robustness has important research significance and application value. Summary of the Invention

[0005] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] To this end, this paper proposes a terrain classification method based on a planar electrical capacitance tomography (ECT) sensor. ECT sensors are a non-contact, real-time, and relatively low-cost measurement technology whose capacitance measurements can reveal the internal dielectric constant distribution of terrain. By integrating a customized planar ECT sensor on the underside of a mobile robot, the method collects capacitance measurement data in real time during navigation, revealing the inherent differences in dielectric constant between different terrain types, thereby achieving efficient and accurate terrain classification.

[0007] To achieve the above objectives, the present invention provides a terrain classification method based on a flat-plate electrical capacitance tomography sensor, comprising:

[0008] Construct a flat-panel ECT sensor array containing eight electrodes;

[0009] Building a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array;

[0010] A terrain proprioception data set is obtained using a mobile robot proprioception data acquisition system. All elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence are extracted according to the corresponding material labels provided in the terrain proprioception data set as the input data of the multi-layer perceptron network. The terrain classification label is used as the output data for network training to obtain a classifier model that can identify different types of terrain. The robot is controlled to move to the terrain to be tested, and the measurement values of inertia data and mutual capacitance data are recorded three times in a preset time. The real-time capacitance time series feature sequence and the real-time inertia time series feature sequence are calculated. The trained classifier model is used to classify the terrain to obtain the terrain classification label of the terrain to be tested to complete the terrain classification.

[0011] The terrain classification method based on the flat-plate electrical capacitance tomography sensor according to the embodiment of the present invention may also have the following additional technical features:

[0012] In one embodiment of the present invention, a flat-panel ECT sensor array comprising eight electrodes is constructed, comprising:

[0013] The eight electrodes of the flat-panel ECT sensor array are arranged in a 2x4 grid and implemented on a PCB. The sensor's external dimensions are 80mm x 85mm. FR-4 material is laid under the sensor plate to isolate external electromagnetic interference. Guard electrodes separate adjacent sensing units, and capacitive coupling is suppressed through a unified ground reference grid.

[0014] In one embodiment of the present invention, a mobile robot proprioception data acquisition system includes: a mobile robot, a multiplexer, a capacitance measurement device, a flat-panel ECT sensor array, a microcomputer, and an inertial measurement unit; wherein,

[0015] The inertial measurement unit is used to measure inertial data;

[0016] The flat-panel ECT sensor array is mounted on the bottom of the mobile robot and connected to a multiplexer. A microcomputer is connected to the multiplexer to control the gating logic. A capacitance measuring device is connected to the multiplexer to collect capacitance data and then save it to the microcomputer. A laptop computer is connected to the microcomputer via a wireless local area network to control the movement of the mobile robot.

[0017] In one embodiment of the present invention, training a multilayer perceptron network includes:

[0018] (3-1) Collecting terrain proprioception dataset:

[0019] (3-1-1) Assume that there are N known terrains in total;

[0020] (3-1-2) The mutual capacitance between the two electrodes collected by the capacitive sensor array in the no-load state is C b =[c1,c2,…,c 28 ], after the mobile robot moves to the terrain O, it continuously collects inertial data and mutual capacitance data, and the mutual capacitance between the two electrodes collected by the capacitance sensor array each time is recorded as C t =[c 1,t ,c 2,t ,…,c 28,t ], the inertial data collected by the inertial measurement unit each time includes the robot's angular velocity ω and acceleration a on the x, y, and z axes, I t =[ω x,t ,ω y,t ,ω z,t ,a x,t ,a y,t ,a z,t ], record the measurement values of three consecutive inertial data C1, C2, C3 and mutual capacitance data I1, I2, I3 within a period of time, and calculate the capacitance time series feature sequence C m =[C1-C b ,C2-C b ,C2-C1,C3-C2] and inertial time series characteristic sequence I m = [I1, I2, I2-I1, I3-I2], complete the acquisition of k capacitance data and inertial data on the terrain O, and obtain k capacitance time series C of the terrain O M,O =[C m,1 ,C m,2 ,…,C m,k ] and k inertial time series I M,O =[I m,1 ,I m,2 ,…,Im,k ], and record the category label of terrain O;

[0021] (3-1-3) Capacitance time series C M,O and inertial time series I M,O For preprocessing, M,O and I M,O All elements in are normalized using the maximum and minimum method;

[0022] (3-1-4) Traverse the N types of terrain in step (3-1-1), repeat step (3-1-2), and obtain the terrain proprioception dataset D IC =[C M,1 ,C M,2 ,…,C M,N ,I M,1 ,I M,2 ,…,I M,N ];

[0023] (3-1-5) Build a multilayer perceptron network. For the 24-dimensional inertial time series data, a two-layer multilayer perceptron network is used, with 22 and 16 neurons in each layer respectively. For the 112-dimensional capacitance time series data, a three-layer MLP is used, with 64, 32, and 16 neurons in each layer respectively. The features extracted from the two types of data are combined into a 32-dimensional vector. The vector is passed through a two-layer multilayer perceptron with 16 and 10 neurons in each layer and a softmax function to calculate the likelihood value of each label. The label with the highest likelihood value is selected as the final terrain classification result.

[0024] (3-2) Terrain proprioception dataset D collected according to step (3-1-4) IC The corresponding material labels provided in (3-1-5) are used as the terrain classifier using the multi-layer perceptron network built as the terrain classifier. M,O All elements C m,i and inertial time series characteristic sequence I M,O All elements in I M,,i As the input of the multi-layer perceptron, the terrain classification label is the output of the multi-layer perceptron, and the network training is performed to obtain a classifier model for identifying terrain types;

[0025] (3-3) After the classifier model is trained in step (3-2), the robot is controlled to move to one of the known terrains O, and the inertial data and mutual capacitance data are measured three times in a period of time, and the corresponding capacitance time series feature sequence C is calculated. m and the corresponding inertial time series characteristic sequence I m, and then use the classifier model to classify the terrain, obtain the terrain classification label of the terrain to be tested O, and complete the terrain classification.

[0026] To achieve the above-mentioned object, the present invention further provides a terrain classification system based on a flat-plate electrical capacitance tomography sensor, comprising:

[0027] A sensor array building module for constructing a flat-panel ECT sensor array containing eight electrodes;

[0028] An acquisition system building module, used to build a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array;

[0029] The classifier model training module is used to obtain a terrain proprioception data set using the mobile robot proprioception data acquisition system, and extract all elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence based on the corresponding material labels provided in the terrain proprioception data set as input data for the multi-layer perceptron network. The terrain classification label is used as output data for network training to obtain a classifier model that can distinguish different terrain types.

[0030] The terrain classification module is used to control the robot to move to the terrain to be tested, record the measurement values of inertia data and mutual capacitance data three times in a preset time, calculate the real-time capacitance time series feature sequence and the real-time inertia time series feature sequence, and use the trained classifier model to classify the terrain to obtain the terrain classification label of the terrain to be tested to complete the terrain classification.

[0031] The terrain classification method and system based on a flat-plate capacitance tomography sensor, according to an embodiment of the present invention, first collects capacitance and inertial data from terrain samples. Then, by constructing and training a multilayer perceptron network, it accurately classifies the terrain samples. Experiments have shown that the system is minimally affected by the robot's motion pattern, achieving excellent classification results for different motion patterns, such as straight lines, curved lines, and those moving at random speeds. Furthermore, the system is unaffected by weather and lighting, enabling all-weather terrain classification.

[0032] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0034] Figure 1 is a flow chart of a terrain classification method based on a flat-plate electrical capacitance tomography sensor according to an embodiment of the present invention;

[0035] Figure 2 is a design diagram of a flat-panel ECT sensor array with eight electrodes according to an embodiment of the present invention;

[0036] Figure 3 is a diagram showing the working principle of capacitive topographic data acquisition according to an embodiment of the present invention;

[0037] Figure 4 is a physical diagram of a mobile robot terrain classification system according to an embodiment of the present invention;

[0038] Figure 5 4 is a structural diagram of a terrain classification system based on a flat-plate electrical capacitance tomography sensor according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0041] The following describes a terrain classification method and system based on a flat-plate electrical capacitance tomography sensor according to an embodiment of the present invention with reference to the accompanying drawings.

[0042] Figure 1 FIG. 1 is a flow chart of a terrain classification method based on a flat-plate capacitance tomography sensor according to an embodiment of the present invention. Figure 1 As shown, the method includes but is not limited to the following steps:

[0043] S1, construction of a flat-plate ECT sensor array containing eight electrodes.

[0044] Specifically, the eight electrodes of the flat-panel ECT sensor array are arranged in a 24-cell grid and implemented as a PCB board. The external dimensions of the sensor are 80mm*85mm. At the same time, to reduce environmental interference (such as external fields) in the planar array, FR-4 material is laid under the sensor plate to isolate external electromagnetic interference. In addition, guard electrodes separate adjacent sensing units, and capacitive coupling is suppressed through a unified ground reference grid.

[0045] S2, building a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array.

[0046] Specifically, the mobile robot proprioception data acquisition system is as follows: Figure 4 As shown, it includes a mobile robot (Agilex LIMO in this embodiment), a multiplexer (MAX306), a capacitance measuring device (Ruby Development Board), a flat ECT sensor array (PCB board design as shown in FIG. Figure 2 As shown in the figure), a microcomputer (Raspberry Pi 4B), an inertial measurement unit (MPU-6050), and a laptop. The working principle diagram of terrain data acquisition is as follows Figure 3 shown.

[0047] Among them, the inertial measurement unit is used to measure inertial data;

[0048] Among them, the flat-panel ECT sensor array is installed on the bottom of the mobile robot and connected to the multiplexer. The microcomputer is connected to the multiplexer to control the selection logic. The capacitance measurement device is connected to the multiplexer to collect capacitance data and then save it to the microcomputer. The laptop computer is connected to the microcomputer through a wireless local area network to control the movement of the mobile robot.

[0049] The flat-panel ECT sensor array of the embodiment of the present invention is mounted on the bottom of the mobile robot, a capacitance measuring device and a multiplexer are mounted on both sides of the mobile robot, a microcomputer is mounted on the top of the mobile robot, and an inertial measurement unit is mounted inside the mobile robot. The capacitance measuring device, the microcomputer, and the multiplexer constitute a capacitance data acquisition unit. The multiplexer is composed of two MAX306 chips. The first eight analog signal input ports CH1-CH8 of each MAX306 are connected to eight channels of the planar array sensor. The address pins A0-A3 of each MAX306 are connected to the microcomputer. The common terminal COM of each MAX306 is connected to the capacitance measuring device. The microcomputer is responsible for controlling the circuit gating between the planar array sensor and the capacitance measuring device to measure the mutual capacitance value between any two of the eight plates.

[0050] The capacitance measurement device includes a capacitance measurement chip, a signal generator, and a main control chip. The signal generator is responsible for generating a periodic excitation voltage. The capacitance measurement chip converts the capacitance value collected by the sensor into 24-bit data and stores it in an internal register. The main control chip reads the data from the register and converts it into hexadecimal. The result is sent as ASCII code in binary form through the serial port and output to the microcomputer. The microcomputer then converts it into the corresponding capacitance value.

[0051] In addition, the microcomputer is also responsible for controlling the movement of the mobile robot. The microcomputer can be connected to the laptop via a network cable and a local area network.

[0052] S3, using the mobile robot proprioception data acquisition system to obtain the terrain proprioception data set, and extracting all elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence according to the corresponding material labels provided in the terrain proprioception data set as the input data of the multi-layer perceptron network, and using the terrain classification label as the output data for network training to obtain a classifier model for identifying different types of terrain; controlling the robot to move to the terrain to be measured, and recording the measurement values of the inertia data and mutual capacitance data three times in a preset time, and calculating the real-time capacitance time series feature sequence and the real-time inertia time series feature sequence, and using the trained classifier model to classify the terrain to obtain the terrain classification label of the terrain to be measured to complete the terrain classification.

[0053] Specifically, to train a network model for terrain classification, the specific steps are as follows:

[0054] (3-1) Collecting proprioceptive data of N known terrains, specifically including the following steps:

[0055] (3-1-1) N=10. The known terrains include: cement roads, grass, stone roads, artificial turf, PVC floors, ceramic tiles, brick roads, parking lot permeable bricks, plastic running tracks, and land.

[0056] (3-1-2) First, measure the mutual capacitance between two electrodes collected by the capacitive sensor array in the no-load state. Specifically, remove the capacitive sensor array at the bottom of the mobile robot and place it flat on the ground, keeping the area above the sensor empty. Measure 200 mutual capacitance values between two electrodes and calculate the average of all values. The mutual capacitance between two electrodes collected by the capacitive sensor array in the no-load state is C. b =[c1,c2,…,c 28], after the mobile robot is controlled to move to the terrain O, the movement modes of the mobile robot include: linear movement (moving forward or backward at a certain speed v, which includes six speeds in this embodiment: v1 = 0.3m / s, v2 = -0.3m / s, v3 = 0.5m / s, v4 = -0.5m / s, v5 = 0.8m / s, v6 = -0.8m / s), curved movement (moving at a combination of a constant linear speed v and an angular speed ω, where v∈{0.3m / s, 0 .5m / s,0.8m / s}, ω∈{0.8rad / s,1.2rad / s}), random motion (the robot generates random linear velocity and angular velocity, where the range of linear velocity v is (0.3m / s,1.2m / s), and the range of angular velocity ω is (0.8rad / s,1.2rad / s)), continuously collects inertial data and mutual capacitance data, the number of electrodes on the electrode plate is 8, and the mutual capacitance between two electrodes collected by the capacitive sensor array each time is recorded as C t =[c 1,t ,c 2,t ,…,c 28,t ], each collected inertial data includes the robot's angular velocity ω and acceleration a on the x, y, and z axes, I t =[ω x,t ,ω y,t ,ω z,t ,a x,t ,a y,t ,a z,t ], the sampling frequency of the capacitance sensor and the inertial sensor are adjusted to 10HZ, and the measurement values of the inertial data C1, C2, C3 and the mutual capacitance data I1, I2, I3 are recorded three times in a period of time, and the capacitance time series feature sequence C is calculated. m =[C1-C b ,C2-C b ,C2-C1,C3-C2] and inertial time series characteristic sequence I m =[I1,I2,I2-I1,I3-I2], the dimensions of the two time series feature sequences are 112 and 24 respectively, k capacitance data and inertial data are collected on the terrain O, and k capacitance time series C of the terrain O are obtained. M,O =[C m,1 ,C m,2 ,…,C m,k ] and k inertial time series I M,O =[I m,1 ,I m,2 ,…,I m,k ], and record the category label of terrain O. In this example, for terrain O, linear motion and curved motion modes are used, and the capacitance time series C obtained for each speed and speed combination is M,O and inertial time series I M,OThe number k is 500, and the random motion mode is used to obtain the capacitance time series C M,O and inertial time series I M,O The number k is 1500, so for the terrain O, after collecting data using all motion modes, a total of 7500 sets (500×6×2+1500) of capacitance time series C are obtained. M,O and inertial time series I M,O ;

[0057] (3-1-3) Capacitance time series C M,O and inertial time series I M,O For preprocessing, M,O and I M,O All elements in are normalized using the maximum and minimum method;

[0058] (3-1-4) Traverse the 10 types of terrain in step (3-1-1), repeat step (3-1-2), and obtain the terrain proprioception dataset D IC =[C M,1 ,C M,2 ,…,C M,N ,I M,1 ,I M,2 ,…,I M,N ];

[0059] (3-1-5) Build a multilayer perceptron network for feature extraction, fusion, and classification of inertial and capacitive data. For 24-dimensional inertial time series data, a two-layer multilayer perceptron network is used, with 22 and 16 neurons in each layer, respectively. For 112-dimensional capacitive time series data, a three-layer MLP is used, with 64, 32, and 16 neurons in each layer, respectively. The features extracted from the two types of data are combined into a 32-dimensional vector. Each layer of neurons contains an activation function. In this example, the ReLU function is used as the activation function. The 32-dimensional vector passes through a two-layer multilayer perceptron, with 16 and 10 neurons in the two layers, respectively, and a softmax function. The likelihood values of 10 terrain labels are calculated, and the label with the highest likelihood value is selected as the final terrain classification result.

[0060] (3-2) Terrain proprioception dataset D collected according to step (3-1-4) IC The corresponding material labels provided in (3-1-5) are used as the terrain classifier using the multi-layer perceptron network built as the terrain classifier. M,O All elements C m,i and inertial time series characteristic sequence I M,O All elements in I M,,iAs the input of the multi-layer perceptron, the terrain classification label is used as the output of the multi-layer perceptron. Training is performed to obtain a classifier that can identify the type of terrain. Specifically, the terrain ontology perception dataset D IC Divide it into training set and validation set, with a ratio of 8:2. IC After randomly shuffling the data in the dataset, the training set and validation set are divided into two parts according to the ratio. Training is started with the batch size set to 64, the maximum number of training epochs to 200, and the learning rate to 0.01. The validation set is used for validation every 10 training rounds. If the classification accuracy is greater than that of the existing model, the model is saved and overwritten (the model is saved directly for the first validation).

[0061] (3-3) After training the classification model according to step (3-2), control the robot to move to one of the 10 known terrains O, record the three measurements of inertial data and mutual capacitance data over a period of time, and calculate the corresponding capacitance time series feature sequence C m and the corresponding inertial time series characteristic sequence I m , and then use the multi-layer perceptron classification model to classify the terrain, obtain the terrain classification label of the terrain to be tested O, and complete the terrain classification.

[0062] The terrain classification method based on the flat-plate capacitance tomography sensor of the embodiment of the present invention designs a flat-plate ECT sensor array containing eight electrodes, and builds a mobile robot body perception data acquisition system to collect the mutual capacitance data and inertial data of the terrain. Secondly, by building a multi-layer perceptron classification model, the capacitance data and inertial data are feature extracted, fused and classified. After the multi-layer perceptron model is trained, the robot is moved to a known terrain, and the terrain classification is completed by the multi-layer perceptron. The present invention utilizes two modal information of inertia and capacitance, and by combining the interactive vibration information between the robot and the terrain and the change of the dielectric constant in the terrain, it can automatically identify terrains of various different materials, and has strong versatility and practical significance.

[0063] In order to implement the above embodiment, Figure 5 As shown, this embodiment also provides a terrain classification system 10 based on a flat-plate electrical capacitance tomography sensor, including:

[0064] A sensor array construction module 100 is used to construct a flat-panel ECT sensor array comprising eight electrodes;

[0065] The acquisition system construction module 200 is used to build a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array;

[0066] The classifier model training module 300 is used to obtain a terrain proprioception data set using the mobile robot proprioception data acquisition system, and extract all elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence based on the corresponding material labels provided in the terrain proprioception data set as input data for the multi-layer perceptron network. The terrain classification labels are used as output data for network training to obtain a classifier model that can distinguish different terrain types.

[0067] The terrain classification module 400 is used to control the robot to move to the terrain to be measured, record the measurement values of inertia data and mutual capacitance data three times in a preset time, calculate the real-time capacitance time series feature sequence and the real-time inertia time series feature sequence, and use the trained classifier model to classify the terrain to obtain the terrain classification label of the terrain to be measured to complete the terrain classification.

[0068] The terrain classification system based on the flat-plate capacitance tomography sensor of the embodiment of the present invention designs a flat-plate ECT sensor array containing eight electrodes, and builds a mobile robot body perception data acquisition system to collect the mutual capacitance data and inertial data of the terrain. Secondly, by building a multi-layer perceptron classification model, the capacitance data and inertial data are feature extracted, fused and classified. After the multi-layer perceptron model is trained, the robot is moved to a known terrain, and the terrain classification is completed by the multi-layer perceptron. The present invention utilizes two modal information of inertia and capacitance, and by combining the interactive vibration information between the robot and the terrain and the change of the dielectric constant in the terrain, it can automatically identify terrains of various different materials, and has strong versatility and practical significance.

[0069] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0070] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

Claims

1. A terrain classification method based on a flat-plate electrical capacitance tomography sensor, characterized in that: include: Construct a flat-panel ECT sensor array containing eight electrodes; Building a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array; A terrain proprioception data set is obtained using a mobile robot proprioception data acquisition system. All elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence are extracted according to the corresponding material labels provided in the terrain proprioception data set as the input data of the multi-layer perceptron network. The terrain classification label is used as the output data for network training to obtain a classifier model that can identify different types of terrain. The robot is controlled to move to the terrain to be tested, and the measurement values of inertia data and mutual capacitance data are recorded three times in a preset time. The real-time capacitance time series feature sequence and the real-time inertia time series feature sequence are calculated. The trained classifier model is used to classify the terrain to obtain the terrain classification label of the terrain to be tested to complete the terrain classification.

2. The method according to claim 1, characterized in that Construct a flat-panel ECT sensor array containing eight electrodes, including: The eight electrodes of the flat-panel ECT sensor array are arranged in a 2x4 grid and implemented on a PCB. The sensor's external dimensions are 80mm x 85mm. FR-4 material is laid under the sensor plate to isolate external electromagnetic interference. Guard electrodes separate adjacent sensing units, and capacitive coupling is suppressed through a unified ground reference grid.

3. The method according to claim 1, characterized in that A mobile robot proprioception data acquisition system includes: a mobile robot, a multiplexer, a capacitance measuring device, a flat-panel ECT sensor array, a microcomputer and an inertial measurement unit; wherein, The inertial measurement unit is used to measure inertial data; The flat-panel ECT sensor array is mounted on the bottom of the mobile robot and connected to a multiplexer. A microcomputer is connected to the multiplexer to control the gating logic. A capacitance measuring device is connected to the multiplexer to collect capacitance data and then save it to the microcomputer. A laptop computer is connected to the microcomputer via a wireless local area network to control the movement of the mobile robot.

4. The method according to claim 1, wherein Training a multilayer perceptron network, including: (3-1) Collecting terrain proprioception dataset: (3-1-1) Assume that there are N known terrains in total; (3-1-2) The mutual capacitance between the two electrodes collected by the capacitive sensor array in the no-load state is C b =[c1,c2,…,c 28 ], after the mobile robot moves to the terrain O, it continuously collects inertial data and mutual capacitance data, and the mutual capacitance between the two electrodes collected by the capacitance sensor array each time is recorded as C t =[c 1,t ,c 2,t ,…,c 28,t ], the inertial data collected by the inertial measurement unit each time includes the robot's angular velocity ω and acceleration a on the x, y, and z axes, I t =[ω x,t ,ω y,t ,ω z,t ,a x,t ,a y,t ,a z,t ], record the measurement values of three consecutive inertial data C1, C2, C3 and mutual capacitance data I1, I2, I3 within a period of time, and calculate the capacitance time series feature sequence C m =[C1-C b ,C2-C b ,C2-C1,C3-C2] and inertial time series characteristic sequence I m = [I1, I2, I2-I1, I3-I2], complete the acquisition of k capacitance data and inertial data on the terrain O, and obtain k capacitance time series C of the terrain O M,O =[C m,1 ,C m,2 ,…,C m,k ] and k inertial time series I M,O =[I m,1 ,I m,2 ,…,I m,k ], and record the category label of terrain O; (3-1-3) Capacitance time series C M,O and inertial time series I M,O For preprocessing, M,O and I M,O All elements in are normalized using the maximum and minimum method; (3-1-4) Traverse the N types of terrain in step (3-1-1), repeat step (3-1-2), and obtain the terrain proprioception dataset D IC =[C M,1 ,C M,2 ,…,C M,N ,I M,1 ,I M,2 ,…,I M,N ]; (3-1-5) Build a multilayer perceptron network. For the 24-dimensional inertial time series data, a two-layer multilayer perceptron network is used, with 22 and 16 neurons in each layer respectively. For the 112-dimensional capacitance time series data, a three-layer MLP is used, with 64, 32, and 16 neurons in each layer respectively. The features extracted from the two types of data are combined into a 32-dimensional vector. The vector is passed through a two-layer multilayer perceptron with 16 and 10 neurons in each layer and a softmax function to calculate the likelihood value of each label. The label with the highest likelihood value is selected as the final terrain classification result. (3-2) Terrain proprioception dataset D collected according to step (3-1-4) IC The corresponding material labels provided in (3-1-5) are used as the terrain classifier using the multi-layer perceptron network built as the terrain classifier. M,O All elements C m,i and inertial time series characteristic sequence I M,O All elements in I M,,i As the input of the multi-layer perceptron, the terrain classification label is the output of the multi-layer perceptron, and the network training is performed to obtain a classifier model for identifying terrain types; (3-3) After the classifier model is trained in step (3-2), the robot is controlled to move to one of the known terrains O, and the inertial data and mutual capacitance data are measured three times in a period of time, and the corresponding capacitance time series feature sequence C is calculated. m and the corresponding inertial time series characteristic sequence I m , and then use the classifier model to classify the terrain, obtain the terrain classification label of the terrain to be tested O, and complete the terrain classification.

5. A terrain classification system based on a flat-plate electrical capacitance tomography sensor, characterized in that: include: A sensor array building module for constructing a flat-panel ECT sensor array containing eight electrodes; An acquisition system building module, used to build a mobile robot proprioception data acquisition system based on the flat-panel ECT sensor array; The classifier model training module is used to obtain a terrain proprioception data set using the mobile robot proprioception data acquisition system, and extract all elements in the capacitance time series feature sequence and all elements in the inertia time series feature sequence based on the corresponding material labels provided in the terrain proprioception data set as input data for the multi-layer perceptron network. The terrain classification label is used as output data for network training to obtain a classifier model that can distinguish different terrain types. The terrain classification module is used to control the robot to move to the terrain to be tested, record the measurement values of inertia data and mutual capacitance data three times in a preset time, calculate the real-time capacitance time series feature sequence and the real-time inertia time series feature sequence, and use the trained classifier model to classify the terrain to obtain the terrain classification label of the terrain to be tested to complete the terrain classification.

Images