基于多层感知器的收获期木薯茎秆识别定位研究

Identification and localization of Cassava Stem Based on Multilayer Perceptron

为解决挖拔式木薯收获机械在作业时需要精确定位茎秆的问题,引入机器视觉技术研究木薯茎秆的识别与定位。以木薯茎秆为研究对象,在室内搭建试验平台模拟木薯收获的田间情况,通过该平台采集模拟的田间图片,对图片预处理得到图像中感兴趣区域;然后,分别提取木薯茎秆和杂物的形状与纹理共9维特征,将其融合作为输入参数,在3层多层感知器网络的结构基础上对参数的选择进行研究,利用主成分分析法对样本特征降维分析,通过交叉验证结合网格搜索的方法试验确定参数最优组合,得出主成分数目为6和隐藏层节点数为9时识别效果最优。基于多层感知器网络构建分类器进行识别效果的定位误差试验,并与支持向量机模型进行了对比,结果表明:MLP分类器识别效果较好,成功率最高达92%,误判率为2%,平均定位误差3.2mm,算法平均耗时0.26s。研究结果可为挖拔式木薯收获机的智能化和自动化作业研究提供参考。

In order to solve the problem that the pulling-up cassava harvester needs to determine the position of the stem during the operation process,machine vision technology is introduced to study the identification and localization of cassava stem.In this paper,cassava stem was selected as the research object and the test platform was built indoors to simulate field trials.Firstly,the acquired image was preprocessed to obtain the region of interest in the image.The 9-dimensional features were extracted from the cassava stems and interferent respectively,and they were combined as input parameters.Then,based on the structure of the three-layer multi-layer perceptron network,the parameter selection was studied.The principal component analysis method was adopted to analyze the dimensionality of the sample features.And the method of cross-validation combined with grid search was used to test and determine the optimal parameters.The recognition effect was optimal when the number of principal components was 6 and the number of hidden layer nodes was 9.Finally,based on the multi-layer perceptron network(MLP)and support vector machine model,the classifier was built to compare the recognition effects and the better classifier was selected for the localization test.The test results showed that the MLP classifier had the best recognition effect,the success rate was 92%,and the error judgment rate was also smaller,which was 2%.The average localization error was 3.2 mm,and the algorithm took an average of 0.26 s.The results demonstrate that the method put forward in the paper has practical application prospects,and provide reference for further application in the pulling-up cassava harvester.