欢迎访问林业科学,今天是

林业科学 ›› 2015, Vol. 51 ›› Issue (7): 129-135.doi: 10.11707/j.1001-7488.20150714

• 问题讨论 • 上一篇    下一篇

基于二维激光与图像的人工林采育目标检测方法

丁小康, 闫磊, 孔建磊, 刘晋浩   

  1. 北京林业大学工学院 北京 100083
  • 收稿日期:2014-07-16 修回日期:2014-10-28 出版日期:2015-07-25 发布日期:2015-08-14
  • 通讯作者: 刘晋浩
  • 基金资助:

    北京高等学校"青年英才计划"(YETP0759); 中国博士后科学基金特别资助项目(2013T60070); 国家林业局引进国际先进林业科学技术(948)项目(2011-4-02)。

A Target Detection Method for Artifical Harvesting Based on 2D Laser and Images

Ding Xiaokang, Yan Lei, Kong Jianlei, Liu Jinhao   

  1. School of Technology, Beijing Forestry University Beijing 100083
  • Received:2014-07-16 Revised:2014-10-28 Online:2015-07-25 Published:2015-08-14

摘要:

[目的] 对人工林区内的采育目标进行检测和识别,为采育作业操作员提供辅助信息,弥补人眼判断的不足,提高作业效率,降低操作风险。[方法] 提出一种基于二维激光和图像的人工林采育目标检测方法,主要内容包括以下几个方面: 1) 基于二维激光测距仪和红外热像仪搭建采育目标信息采集系统,利用上位机对传感器的信息采集进行控制,并对采集到的信号进行预处理,获取目标的激光数据、可见光图像和红外热图像; 2) 将激光与图像进行标定,得到图像中的目标区域,同时由激光坐标得到目标的位置信息,为目标识别和定位打好基础; 3) 将可见光图像与红外图像进行融合,融合后的图像信息更丰富并减少由单一传感器所引起的不确定性,起到互补的作用; 4) 根据采育目标的特点,基于激光与图像信息进行特征提取,包括温度特征、颜色特征和形状特征等,为目标识别提供具体依据; 5) 在获得采育目标特征的基础上,运用当前流行的机器学习算法——支持向量机(SVM),通过大量样本训练和学习,建立特征参数与采育目标的数学模型; 6) 在SVM模型的基础上,利用3种不同的优化算法对其参数进行优化,提高识别性能。[结果] 将不同算法优化后的模型通过受试者工作特征曲线(ROC曲线)进行对比,结果显示,采用遗传算法优化后的模型对人工林采育目标的识别正确率能够达到96%以上,具有较好的识别效果。[结论] 将多传感器融合技术引入到林业智能装备中,取代常用的昂贵的三维激光扫描系统,采用二维激光与图像结合的方式,一方面节省了成本,另一方面,针对二维数据,系统的数据处理速度更快;同时,多传感器之间的互补作用使得对目标的测量和识别更为准确。

关键词: 激光, 图像, 采育目标, 识别, 检测

Abstract:

[Objective] When large forestry equipment are in practice, it needs operator to make a lot of observation to determine due to the complexity of forest environments and the impact of obstacles, which result in intermittent operation, reducing operational efficiency. Especially in poor light conditions, it will greatly increase the risks for the operator. Based on 2D laser and images, harvesting targets in artificial forest were detected and identified in this study by using information fusion technology, which provided supplementary information to the operator to compensate for the lack of the human eyes. [Method] This paper presented a target detection method for artificial forest harvesting based on 2D laser and images. The main contents were: 1) Built harvesting target information collection system based on 2D laser scanner and thermal imager, where PC was applied to control the information collection. Also the collected signal pretreatment was conducted in this part. Then laser data, visible and infrared images were captured for targets. 2) Internal and external joint calibration was conducted between laser points and image to match them together, thereby obtaining the target area image we need at the same time to obtain the coordinate position of the target by the laser information, which provided the basis for subsequent target recognition. 3) Fused the visible image and infrared image. Fused image had richer information and reduced uncertainty caused by a single sensor. 4) According to the collected information, features including temperature, colors and shape were extracted to provide specific evidence for target identification. 5) On the basis of obtained features, a popular machine learning algorithms-support vector machineswas applied by training and learning with a number of samples to establish math model for features and targets. 6) The characteristic parameters of the mathematical model were optimized by 3 different optimization algorithms to improve recognition performance. [Result] Models with different optimizations were compared by ROC curves. Experiment results showed that the optimized model by GA used in this paper could recognize harvesting targets effectively and it’s correct rate could reach more than 96%. [Conclusion] Multi-sensor fusion technology was introduced into smart forestry equipment in this paper. Instead of the commonly used expensive 3D laser scanning system, 2D laser and images were combined and used. On the one hand, it saved the costs; on the other hand, for the 2D data, the data processing was faster. At the same time, the complementary among sensorsmade the measuring and identification more accurate. The results showed that the method used in this study could provide the information needed by forestry equipment operations, which improved operational efficiency and reduced operational risks.

Key words: laser, image, harvesting target, recognition, detection

中图分类号: