基于机器学习的无人船目标识别系统研究

Unmanned Ship Target Recognition System Based on Machine Learning

针对无人船装置在复杂视域环境下的目标识别准确率低、运算率高等不足,提出了采用OpenCV视觉处理框架,建立基于机器学习的无人船目标识别系统。利用HAAR级联分类器训练算法建立了目标物的机器学习库,通过多组训练目标物对比实验,分析出最大识别率和最小虚警率的关系,及正负样本尺寸和比例对训练时间和精确度的影响,得出参数值的最适设定范围。还通过水面的镜面效应,采用相位相关性法水岸线识别算法,准确地识别出水岸线,提高了水中目标物的识别效率。

In oder to solve the problems of the low accuracy of target recognition and high calculation rate of the unmanned ship device in the complex field of view environment,the OpenCV visual processing framework is proposed to establish a unmanned ship target recognition system based on machine learning.The HAAR cascade classifier training algorithm was used to establish the machine learning library of the target object.Through the comparison experiments of multiple groups of training objects,the relationship between the maximum recognition rate and the minimum false alarm rate,and the influence of the positive and negative sample size and proportion on training time and accuracy were analyzed,and the optimum setting range of parameter values was obtained.Through the mirror effect of the water surface,the phase correlation method waterfront recognition algorithm was used to accurately identify the water shoreline and improve the recognition efficiency of the target in the water.