基于机器视觉和GABP算法的播种检测研究

Research on Prediction of Seed Number in the Hole Based on Machine Vision and GABP Algorithm

为了判别育秧播种后每穴超级稻籽粒的数目,基于机器视觉运用了遗传改进的BP神经网络(GABP)算法并结合LabVIEW和Matlab设计了图像采集、处理及结果显示系统。针对采集到的图像,采用迭代阈值法分割并获取二值图像,运用投影法定位秧盘目标检测区域及秧穴,并提取超级稻的面积、周长、形状因子和7个不变矩共10个特征参数,建立基于GABP算法的播种检测模型,分别检测空穴、1粒、2粒、3粒。检测试验结果表明,4种情况的实际检测试验结果与人工检测相对误差分别为3.9%、2.0%、3.74%、5.63%,算法平均处理时间为1.018 s,为进一步实现在线穴播量检测系统设计及精密播种作业提供参考。

In order to judge the seed number in each hole, the system used for image capture, processing, and display the prediction results is designed based on machine vision using GABP algorithm and combining with LabVIEW and Matlab. Iterative threshold method is used to partition and get binary image, projection method is used to locate target area and extact ten morphological characteristics including area, perimeter, shape factor of super rice and seven invariant moments of connected domain. The seeding detection model based on GABP al- gorithm is established for estimating seed numbers in each hole. The detection experiment results show that rel- ative error rates of the actual detection and manual detection are 3.9%, 2.0%, 3.74% and 5.63%, respec- tively, corresponding to 0, 1,2, 3 seeds in each hole, and the average processing time is about 1. 018 s. All the results and methods provide a reference for the design of the online prediction system and further realization of the precision planting.