间苗是保证直播油菜增产的一项关键技术措施,为解决人工间苗劳动强度大、机械间苗不精确的问题,采用机器视觉的方式,基于深度学习算法YOLO v5平台,设计并搭建自动间苗装置.机器视觉系统评估幼苗种群的合理密植情况,间苗算法以间距和幼苗叶展为评估标准,实现控制幼苗间距和筛选优质苗的功能.选用遗传算法对间苗行进路径进行规划,相较于未规划路径可缩短最低为50％的行进距离,最终采用激光器高温烧除的方式完成间苗作业.选取油菜苗作为试验材料,间苗阈值α是划定幼苗最小间距的参数,设置不同的间苗阈值α进行试验.结果表明,间除苗的数量随着间苗阈值α的增加而增加,幼苗平均密度下降的同时种群分布趋于均匀,对间除苗的叶展长度分类统计,α为0～75 mm时,间除苗叶展长度全部在0～20 mm范围;α为75～200 mm时,间除苗叶展长度为0～40 mm,其中叶展长度为20～40 mm的最高占比约为76％;α为200～350 mm时,间除苗叶展长度在40 mm以上的幼苗开始增加,最高占比约为14％,间除苗叶展长度梯次分布证明了间苗算法具备筛选优质苗的性能.间苗执行阶段耗时占据间苗作业总耗时的90％以上,以激光走线参数L、激光器功率P、间苗距离阈值为试验因素,三因素三水平正交试验结果表明:选择合适的激光走线参数L能有效提高间苗死亡率、降低间苗误伤率和减少间苗耗时,在参数L为30mm、P为7.5W、α为250 mm下开展土槽台架性能验证试验,激光间苗平均死亡率为93.29％,平均误伤率为5.19％,平均总耗时为15.19 min,为开发基于机器视觉的激光自动间苗机提供了理论基础和技术支撑.
Seedling is a key technical measure to ensure the increase of direct canola yield,based on the deep learning algorithm YOLO v5 detection platform,a visual seedling system was developed,which can visualize the planting density of seedlings and help growers grasp the situation of reasonable dense planting.An interseedling algorithm with spacing and seedling leaf spread as the evaluation criteria was developed,which can standardize the seedling spacing and meet the horticultural requirements of seedlings to remove inferiority and store excellent seedlings.The genetic algorithm was used to plan the travel path of seedlings to improve the efficiency of seedlings.Compared with unplanned routes,the travel distance can be shortened by more than 50％.Rapeseed seeds were selected as experimental materials,and set different threshold values α for the experiment.The results showed that with the increase of the threshold α of inter-seedlings,the number of inter-seedlings was also increased,and the average density of the identified area was decreased,and the leaf spread length of the inter-seedlings was classified and counted,the ladder distribution of the leaf length of the seedlings proved that the seedling extension algorithm had the function of eliminating inferiority and preserving excellence.The execution stage of the seedlings accounted for more than 90％of the total time spent in the seedling operation.The three-factor three-level orthogonal test was carried out with the laser trace range,laser power,distance threshold,as the test factors.The performance verification test was carried out under the parameters of the optimal interseedling L of 30 mm,the laser power P of 7.5 W,and the inter-seedling threshold α of 250 mm,reducing the accidental injury rate of seedlings and the time consumption of seedlings.The experiment showed that the average mortality rate of laser seedlings was 93.29％,the average false injury rate was 5.19％,the average total time spent was 15.19 min.The research result provided a theoretical basis for automatic seedlings and provided equipment support for agricultural industrial planting.
张昌松;李伟
陕西科技大学机电工程学院,西安 710021
农业工程
油菜;自动间苗;YOLO v5;激光器;遗传算法
rape;automatic seedlings;YOLO v5;laser;genetic algorithms
《农业机械学报》 2024 (004)
40-52 / 13
西安近代化学研究所开放合作创新基金项目(SYJJ200304)
10.6041/j.issn.1000-1298.2024.04.004
评论