FMCW LiDAR FMCW LiDAR SIGNAL PROCESSING SYSTEM AND METHOD BASED ON NEURAL NETWORK

摘要

The invention relates to a FMCW lidar signal processing system and method based on a new network.More specifically, set a loose margin for the average clipping factor of CFAR (constant false alarm rate), use neural network to detect all meaningful data in the output data, and obtain additional data from the optical community that could not be obtained before. In addition,The peak of FFT (fast Fourier transform) results can be interpolated at a higher resolution than linear interpolation due to many weights of neural network.Signal attenuation proportional to distance,Alternatively, NN (neural network) can be implemented to realize multi-objective beat frequency detection,This is about the implementation of NPU (neural processing unit), which can accelerate the action in one cycle of FFT (fast Fourier transform). For this purpose, the present invention,Loading pre saved preferences or new settings when FMCW lidar is started or reset, and loading steps and;A control signal generation stage and for generating a digital laser control signal according to the value set in the setting stage;After the above control signal generation stage, receiving the emitted laser and converting it into a frequency domain conversion stage and;The frequency domain converted through the above frequency conversion stage searches the candidate region search stage and of the effective data candidate region through the CFAR filter;Performing Accelerated neural network calculation on the effective data candidate area searched from the candidate area search stage, and classifying the effective data in the effective data classification stage and;An effective data storage stage for storing the classified effective data after the effective data classification stage;Characterized by inclusion. In addition,The invention is,A laser receiving part that receives the reflected laser and;A signal conversion unit that converts the laser received by the laser receiving unit into a digital signal, and;The signal conversion unit converts the received digital signal into a frequency conversion unit in the frequency domain and;The frequency domain converted by the frequency conversion unit passes through the CFAR filter to search the candidate region search unit of the effective data candidate region and;An effective data classification unit that performs accelerated neural network calculation on the effective data candidate region searched by the candidate region search unit and classifies the effective data;An effective data storage book storing the effective data classified by the above effective data classification book and;Based on the data stored in the effective data storage unit, set the variables of the control signal processing system and set the system control unit for monitoring;Characterized by inclusion.