番茄采摘机器人非颜色编码化目标识别算法研究

为了实现番茄采摘机器人在非结构化环境下对目标番茄的准确识别,提出了一种基于非颜色编码的番茄识别算法。通过Haar-like特征及其编码的方法,结合AdaBoost深度学习算法可以获得用于识别成熟番茄的分类器;并研究了Haar-like特征类型和AdaBoost学习训练次数对分类器性能的影响。所得强分类器对测试集中的番茄进行在线识别试验。试验结果表明,测试集中93.3%的成熟番茄能够被正确识别;同时该分类器还对光照变化、果实粘连以及枝叶遮挡等干扰具有较强的自适应性和鲁棒性,满足采摘机器人对目标识别的技术要求。

Object Recognition Algorithm of Tomato Harvesting Robot Using Non-color Coding Approach

In order to detect the ripe tomato in unstructured environment for robotic harvesting, a tomato recognition algorithm using non color coding approach was developed. The proposed algorithm was consist of offline training and online recognition. In the process of offline training, a strong classifier was obtained using AdaBoost algorithm with Haar-like features. The Haar-like feature is a kind of non color coding feature which can be extracted by integral figure calculation. In the online recognition process, the tomato object was detected by using the strong classifier which was obtained in the offline training process. Two couples of comparative tests were conducted to study the influence of the types of Haar-like features and training times on the performance of the proposed algorithm. The results showed that the C style Haar-like features and 20000 training times were the optimal parameters for the size of training set. The results of online recognition tests indicated that about 93.3% ripe tomatoes existing in the testing samples set were successfully detected. The proposed tomato recognition approach was also successfully applied in the unstructured environment with various disturbances such as occluded, overlapping, and varying illumination, which indicated that the proposed tomato recognition algorithm was self adaptive and robust. It was available to be applied in the vision recognition system for a harvesting robot.