Navigation method of intelligent pseudo-ginseng harvester based on LSTM and GNSS / INS multi-mode fusion

Abstract

The invention discloses a navigation method of an intelligent pseudo-ginseng harvester based on LSTM and GNSS / INS multi-modal fusion, which comprises the following steps: firstly, acquiring multi-modal data of the harvester through a GNSS receiver and an INS inertial measurement unit, and performing timestamp alignment and noise suppression preprocessing on the multi-modal data; then, a GNSS / INS data fusion model is constructed, preliminary data fusion is realized through a Kalman filtering algorithm, and the positioning precision is improved; meanwhile, a long short-term memory network (LSTM) model is introduced, during the interruption period of GNSS signals, historical INS and GNSS data are utilized to predict GNSS pseudo position increment, pseudo GNSS position information is generated, and the pseudo GNSS position information and INS resolving data are fused again; and finally, in combination with navigation information fused by a Kalman filtering algorithm, realizing accurate positioning of the pseudo-ginseng harvester. The method can effectively improve the precision and robustness of a navigation system, and is suitable for efficient operation of the pseudo-ginseng combine harvester in complex environments of hills and mountains.

Description

Technical Field

[0001] The invention relates to a navigation method of an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multi-modal fusion, and belongs to the field of intelligent agricultural machinery. Background Art

[0002] With the development of modern agriculture, the demand for the application of precision agricultural mechanized operation technology in complex terrain conditions is increasing. As a traditional Chinese medicinal material, Panax notoginseng is mostly planted in hilly and mountainous areas with complex terrain and large undulations. Traditional manual or mechanical harvesting methods are difficult to meet the requirements of efficient and precise operations. Therefore, intelligent navigation technology has become an important technical support for Panax notoginseng combine harvesters.

[0003] However, the current navigation technology still has shortcomings in hilly and mountainous environments: on the one hand, the global satellite navigation system (GNSS), as the mainstream navigation technology, can provide global position information, but it is susceptible to occlusion and multipath effects in hilly and mountainous environments, resulting in unstable or even interrupted signals, causing large positioning errors in the navigation system. On the other hand, the inertial navigation system (INS) can provide high-frequency acceleration and angular velocity information through inertial sensors, and has the ability to navigate independently in a short period of time, but its positioning accuracy accumulates and drifts over time, and long-term use will cause error accumulation problems. The current GNSS / INS combined navigation method attempts to alleviate the above problems through fusion algorithms. However, these methods still have shortcomings in complex dynamic environments, resulting in the navigation accuracy and robustness of the system under dynamic changes and occlusion conditions that are difficult to meet actual needs.

[0004] To address the above problems, this paper proposes a navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion. By introducing deep learning technology to perform short-term prediction of GNSS signals and combining it with INS data, the problem of navigation accuracy drifting over time is solved. At the same time, the robustness and continuity of the system in complex terrain are improved, providing efficient and accurate navigation support for the Panax notoginseng combine harvester. Summary of the invention

[0005] Aiming at the problems that global satellite navigation system (GNSS) signals are susceptible to occlusion, multipath interference and accumulation of inertial navigation system (INS) drift errors in hilly and mountainous environments, an intelligent Panax notoginseng harvester navigation method based on long short-term memory network (LSTM) and GNSS / INS multimodal fusion is proposed, which can complete the precise navigation of Panax notoginseng combine harvester.

[0006] The present invention provides a navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion, comprising the following steps: A navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion includes the following steps: S1. Multimodal data acquisition and preprocessing: Coordinates are collected through GNSS module and INS module respectively ,speed , acceleration and angular velocity Data, denoise and align timestamps of the collected raw data to ensure the consistency of multimodal data at the same time, define k The state vector at time ; S2. GNSS / INS data fusion model construction: Use Kalman filter algorithm to fuse GNSS and INS data; S201: Predicted state estimation: Use the state at the previous moment and the acceleration and angular velocity measured by INS to predict the current position and velocity; S202: Prediction error covariance: Update the covariance matrix of the state estimate based on the system model and sensor noise to reflect the error accumulation in the prediction stage; S203: Calculate Kalman gain: Calculate Kalman gain by combining the prediction error covariance and the GNSS positioning error to determine the trust ratio between the prediction value and the measured value; S204: Update state estimation: Use the coordinate and velocity measurement data provided by GNSS to correct the predicted value through Kalman gain to obtain a more accurate state estimation result; S205: Update error covariance: adjust the covariance matrix according to the Kalman gain to reduce the uncertainty of state estimation and provide a basis for filtering at the next moment; Through the above steps, the position, velocity, acceleration and angular velocity are accurately estimated through continuous iteration; S3. GNSS data prediction based on LSTM: In the case of GNSS signal interruption or interference, the LSTM model is used to predict the pseudo-GNSS position increment. The input of LSTM is a combination of historical INS and GNSS data: S301, input layer: The input of the LSTM network is ; in n represents the length of the time window, t is the current time step, S302, Hidden State and Memory Unit: LSTM extracts the temporal correlation of input data through its internal structure: Forget Gate: in: and is the weight and bias of the forget gate, is the hidden state at the previous time step, is the input of the current time step, is the activation function, Input Gate: Update memory unit: Output Gate: Update the hidden state: S303, output layer: The output of the LSTM network is the GNSS pseudo position increment of the current time step , realized through linear mapping: in, and are the weights and biases of the output layer; S304, Time Series Modeling Process: LSTM network uses multiple time-step input sequences Extracting temporal correlation, the model captures the long-term dependency between GNSS and INS through a recursive structure, thereby accurately predicting the pseudo-position increment when the GNSS signal is interrupted; S4. GNSS signal interruption model output: When GNSS signal interruption is detected, the model output Indicates the change in GNSS position at the current moment, by accumulating the position at the previous moment Calculate the current pseudo GNSS position ; S401, multi-modal data fusion and navigation update: input the pseudo GNSS position information and INS solution data into the Kalman filter algorithm again for data fusion, update the state variables, and generate high-precision navigation information for the harvester; S5. Output navigation position information: The fused navigation data output includes current position information, speed information and attitude information.

[0007] During the training process, the mean square error (MSE) is used as the loss function To ensure the stability of the algorithm and navigation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flowchart of a multi-modal fusion method based on LSTM and GNSS / INS in an embodiment of the present invention; Figure 2 is a flow chart of GNSS / INS multimodal fusion in an embodiment of the present invention; Figure 3 LSTM data training structure diagram in an embodiment of the present invention; Figure 4 LSTM network structure diagram in an embodiment of the present invention; Figure 5 This is a block diagram of the LSTM data compensation structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments: The present invention provides a navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion. By introducing LSTM, a model capable of predicting pseudo-position increments during GNSS signal interruption is constructed, so that the system has the ability to continuously provide high-precision navigation information based on historical data. The multimodal data fusion framework adopts the Kalman filter algorithm to achieve dynamic fusion of GNSS and INS data under both normal signal and pseudo-GNSS signal conditions. In addition, the method designs a set of time series modeling mechanisms to ensure the stability and accuracy of the prediction by recursively capturing the long-term correlation between GNSS and INS data.

[0010] like Figure 1 As shown, one embodiment of the present invention discloses a navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion, comprising the following steps: S1. Multimodal data acquisition and preprocessing: Coordinates are collected through GNSS module and INS module respectively ,speed , acceleration and angular velocity Data, denoise and align timestamps of the collected raw data to ensure the consistency of multimodal data at the same time, define k The state vector at time ; S2. GNSS / INS data fusion model construction: Use Kalman filter algorithm to fuse GNSS and INS data; like Figure 2 As shown in the figure, the steps of GNSS / INS multimodal fusion are as follows: S201: Predicted state estimation: Use the state at the previous moment and the acceleration and angular velocity measured by INS to predict the current position and velocity; S202: Prediction error covariance: Update the covariance matrix of the state estimate based on the system model and sensor noise to reflect the error accumulation in the prediction stage; S203: Calculate Kalman gain: Calculate Kalman gain by combining the prediction error covariance and the GNSS positioning error to determine the trust ratio between the prediction value and the measured value; S204: Update state estimation: Use the measured data such as coordinates and speed provided by GNSS to correct the predicted value through Kalman gain to obtain a more accurate state estimation result; S205: Update error covariance: adjust the covariance matrix according to the Kalman gain to reduce the uncertainty of state estimation and provide a basis for filtering at the next moment; Through the above steps, the position, velocity, acceleration and angular velocity are accurately estimated through continuous iteration; S3. GNSS data prediction based on LSTM: In the case of GNSS signal interruption or interference, the LSTM model is used to predict the pseudo-GNSS position increment. The input of LSTM is a combination of historical INS and GNSS data: like Figure 3 As shown, the LSTM data training steps are as follows: S301, input layer: The input of the LSTM network is ; in n represents the length of the time window, t is the current time step, S302, Hidden State and Memory Unit: LSTM extracts the temporal correlation of input data through its internal structure: like Figure 4 As shown, the LSTM network steps are as follows: Forget Gate: in: and is the weight and bias of the forget gate, is the hidden state at the previous time step, is the input of the current time step, is the activation function, Input Gate: Update memory unit: Output Gate: Update the hidden state: S303, output layer: The output of the LSTM network is the GNSS pseudo position increment of the current time step , realized through linear mapping: in, and are the weights and biases of the output layer; S304, Time Series Modeling Process: LSTM network uses multiple time-step input sequences Extracting temporal correlation, the model captures the long-term dependency between GNSS and INS through a recursive structure, thereby accurately predicting the pseudo-position increment when the GNSS signal is interrupted; like Figure 5 As shown in: S4, GNSS signal interruption model output: When the GNSS signal is interrupted, the model output Indicates the change in GNSS position at the current moment, by accumulating the position at the previous moment Calculate the current pseudo GNSS position ; S401, multi-modal data fusion and navigation update: input the pseudo GNSS position information and INS solution data into the Kalman filter algorithm again for data fusion, update the state variables, and generate high-precision navigation information for the harvester; S5. Output navigation position information: The fused navigation data output includes current position information, speed information and attitude information.

[0011] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A navigation method for an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion, characterized in that: The steps include: S1. Multimodal data collection and preprocessing: Coordinate collection through GNSS module and speed Data, INS module collects acceleration and angular velocity Data, denoise and align timestamps of the collected raw data to ensure the consistency of multimodal data at the same time, define k The state vector at time ; S2. GNSS / INS data fusion model construction: Use Kalman filter algorithm to fuse GNSS and INS data; S201: Predicted state estimation: Use the state at the previous moment and the acceleration and angular velocity measured by INS to predict the current position and velocity; S202: Prediction error covariance: Update the covariance matrix of the state estimate based on the system model and sensor noise to reflect the error accumulation in the prediction stage; S203: Calculate Kalman gain: Calculate Kalman gain by combining the prediction error covariance and the GNSS positioning error to determine the trust ratio between the prediction value and the measured value; S204: Update state estimation: Use the coordinate and velocity measurement data provided by GNSS to correct the predicted value through Kalman gain to obtain a more accurate state estimation result; S205: Update error covariance: adjust the covariance matrix according to the Kalman gain to reduce the uncertainty of state estimation and provide a basis for filtering at the next moment; Through the above steps, the position, velocity, acceleration and angular velocity are accurately estimated through continuous iteration; S3. GNSS data prediction based on LSTM: In the case of GNSS signal interruption or interference, the LSTM model is used to predict the pseudo-GNSS position increment. The input of LSTM is a combination of historical INS and GNSS data: S301, input layer: The input of the LSTM network is ; in n represents the length of the time window, t is the current time step, S302, Hidden State and Memory Unit: LSTM extracts the temporal correlation of input data through its internal structure: Forget Gate: in: and is the weight and bias of the forget gate, is the hidden state at the previous time step, is the input of the current time step, is the activation function, Input Gate: Update memory unit: Output Gate: Update the hidden state: S303, output layer: The output of the LSTM network is the GNSS pseudo position increment of the current time step , realized through linear mapping: in, and are the weights and biases of the output layer; S304, Time Series Modeling Process: LSTM network uses multiple time-step input sequences Extracting temporal correlation, the model captures the long-term dependency between GNSS and INS through a recursive structure, thereby accurately predicting the pseudo-position increment when the GNSS signal is interrupted; S4. GNSS signal interruption model output: When GNSS signal interruption is detected, the model output Indicates the change in GNSS position at the current moment, by accumulating the position at the previous moment Calculate the current pseudo GNSS position ; S401, multi-modal data fusion and navigation update: input the pseudo GNSS position information and INS solution data into the Kalman filter algorithm again for data fusion, update the state variables, and generate high-precision navigation information for the harvester; S5. Output navigation position information: The fused navigation data output includes current position information, speed information and attitude information.

2. The navigation method of an intelligent Panax notoginseng harvester based on LSTM and GNSS / INS multimodal fusion according to claim 1, characterized in that: During the training process, MSE is used as the loss function To ensure the stability of the algorithm and navigation accuracy.

Images