APD Bias Voltage Control Method Based On Improved Random Forest

Light Detection and Ranging(Li DAR)is a high-precision ranging technology widely used in autonomous driving,intelligent transportation,robots,unmanned aerial vehicles and other fields.As one of the core detection components of array Li DAR,the focal plane detector has the advantages of high sensitivity and low noise,which can realize the efficient detection and recognition of target objects in complex environments.However,in the working process of the Avalanche Photodiode(APD)of the focal plane detector,the bias voltage has an important impact on its performance.If the bias voltage is too low,the detection ability of the detector is weak,and when the bias voltage is too high,it is easy to damage the detector.Therefore,research on how to control the bias voltage of APD to improve the measurement accuracy and stability of Li DAR has become one of the current research hotspots.In this paper,the Li DAR system is designed and implemented,and the control of APD bias voltage in the avalanche focal plane detector is studied.The specific content is as follows:(1)Design and implementation of Li DAR system.In this paper,the composition and working principle of the Li DAR system are introduced,and the parameters of the core components such as laser,avalanche focal plane camera and main control chip are introduced.The experimental equipment is selected and developed,the operating system is transplanted and the system control software is written to control the Li DAR system,data storage and transmission.Finally,the basic functions are tested in the optical laboratory,which basically meets the design requirements of the system in this paper and achieves the expected level.(2)Research on APD bias voltage control method based on improved random forest.In this paper,the range profile data of APD output is preprocessed and then feature extraction is performed.In the algorithm model,the traditional random forest algorithm was improved,that is,when the decision tree was voting,the classification effect of the out-of-bag data in each training subset was used as the weight of the tree.In order to solve the problem of difficult parameter selection,the particle swarm optimization algorithm is used to iteratively optimize the parameters that affect the new model.Finally,the biasing voltage of the APD is adjusted by the dichotomy compensation method until the APD is in the best working state.(3)Experiment of APD bias voltage control method based on improved random forest.By using different lengths of fiber to simulate avalanche focal plane detectors to detect objects at different distances,the classification effects of improved random forest algorithm,support vector machine,random forest and other algorithms in processing range profile data were compared.The results show that the classification accuracy of the improved random forest reaches more than 98%,which is better than other algorithms.Finally,by comparing the APD bias voltage control method based on random forest and the improved random forest,the evaluation indexes show that the structural similarity of the range profile is increased by2%～7%,and the root mean square error is decreased by 4% ～7%,and the quality of the range profile is improved.The results show that the improved random forest algorithm can accurately judge the working state of APD compared with the traditional machine learning algorithm,and the algorithm model can be used to control the APD bias voltage.The experimental results show that the proposed control method can adaptively control the APD bias voltage of the avalanche focal plane detector according to the change of external environmental factors,which can make the avalanche focal plane camera work in the best condition,and improve the stability and reliability of the Li DAR system.This study provides a new idea and method for machine learning in APD bias voltage control.