尺度分层多阈值结合SVM分类的热红外舰船识别

针对尺度自适应选择分层多阈值方法,有时检测目标不完整且存在较多虚警目标等问题,提出了一种基于尺度分层多阈值和SVM分类相结合的舰船目标检测与识别方法。首先使用尺度自适应分层多阈值方法进行初检测;其次根据样本学习生成舰船目标特征及最佳分类特征组合;最后使用SVM样本学习分类实现舰船目标检测与识别。实验结果表明,该方法比单一使用样本分类法降低了虚警率,提高了检测效率,能在近岸舰船目标与背景对比度较低的情况下实现虚假目标有效剔除,且在突堤式码头停放的舰船目标识别中更有效和更稳定。

Recognition of Thermal Infrared Ships Targets Based on Scale Hierarchical Multi-Threshold and SVM Classification

A method for detecting ship targets that combines improved scale hierarchical multi-threshold and SVM sample learning is proposed due to incomplete and many false alarm targets for the approach of hierarchical multithreshold detection based on scale-adaptive selection.First of all,initial detection was conducted on the basis of improved scale-adaptive selection and hierarchical multi-threshold.Then the features of ship targets and an optimal combination of classification features were generated from sample learning.Finally,ship targets were detected through classification based on SVM sample learning.The results of the experiment show this method reduces time consumption of the detection than that of the single sample classification and this method can achieve the effective elimination of false targets in the case of the low contrast between the target and the background of the offshore ship targets.This method is capable of more stable and efficient recognition to ship targets parked in jetty type wharf.