Abstract: In order to solve the problem of low sorting efficiency after picking and shelling of Camellia fruits in machinevision, the paper proposes a preference artificial immune network algorithm (aiNet) with multifeatures, which applies the multiobjective optimization of artificial immune network and preference database to extract the multifeatures in color and shape of Camellia objects to input into the immune network for simulation test. The test results show that the multifeatures preference immune network proposed in this paper is feasible. The recognition rate of the network has reached 90%, and the minimum identification time is 60 ms. Compared with the single feature sorting method, this method is more effective, which provides a feasible scheme for the intelligent sorting method of agricultural and forestry targets.

Key words: sorting, machinevision, Camellia fruit, aiNet, multifeatures

摘要： 针对油茶果采摘、脱壳后机器视觉分选效率不高的问题,提出一种多特征偏好人工免疫网络算法,该算法应用人工免疫网络的多目标优化与偏好数据库特征,提取油茶目标的颜色、形态多特征输入免疫网络进行仿真测试。试验结果表明,本文提出的多特征偏好免疫网络的识别率最高达到90%以上。相比单特征分选方法有了较大的提升,证明本文分选方法的有效性,并为农林业目标智能化分选辨识提供一种可行的方案。

关键词: 分选, 机器视觉, 油茶果, 人工免疫网络, 多特征