基于机器视觉的超级杂交稻秧盘育秧智能播种新技术研究

Super hybrid rice is a kind of rice variety that provides excellent quality and high yield. Super hybrid rice is very popular as a large-scale rice crop in China.The precision seeding of tray nursing seedling requires 2 ± 1 seeds per hole. At present, the performance of metering devices can reach 1-3 seeds per hole, but the rate of single seed is high (more than 20%), and there are cavities. Due to factors such as the germination rate, survival rate, characteristics of blanket and injured seedling rate, the seeding results can not reach 1-2 planting per hole. Therefore, seeding target should be raised to 2-3 seeds per hole, which will reduce single-grain rate and eliminate the presence of cavities. This project focuses on imitating the manual sowing principle by monocular camera as eyes, the vision inspection systems as the brain, and the hill-drop device performing hand movements. This innovative intelligent seeding technology will be put forward based on machine vision. First, the features of the seed image and kinematics characteristics of the required seed vacuum suction will be studied, and the seed model database of image features will be established. The boundary conditions of seed vacuum suction will be obtained, the images of the seedling trays will be detected dynamically using the fuzzy analysis method. The seeding programs will be implemented according to the seeding target. The intelligent hill-drop machine will be designed with umbrella shaped suction needles and a vibration type filling device, with a trapezoidal seed circulation flow. An algorithm of point-position motion control will be studied. The seeding track will be optimized with the suction seed boundary conditions and efficiency.The proposed project will be able to achieve an idea of seeding and supplemental seeding according to the required number of seeds per hole. Research method will open up a new precision seeding technology, raise the level of planting accuracy，and have important scientific significance and application prospects.

超级杂交稻是品质优良、产量高的新型水稻品种，已经在我国大面积推广应用，其秧盘育秧精密播种要求2±1粒/穴，目前排种器的性能可以达到1～3粒/穴，但单粒率偏高（20%以上），并存在空穴，因受发芽率、成苗率、成毯性及伤秧率等影响，结果不能达到1～2株/穴种植要求，因此播种目标应提升为2～3粒/穴，且单粒率少、无空穴。本项目模仿人工播种原理，以单目摄像机作为眼睛，视觉检测系统作为大脑，点播器执行手的播种动作，创新性提出一种基于机器视觉的智能播种技术。首先研究种子图像特征和吸种运动学特性，建立种子图像特征模型库，确定吸种边界条件，对秧盘图像进行动态检测及模糊分析，按目标规划播种方案，设计具有伞式多粒吸种和循环流动振动充种的智能点播机，研究点位运动控制算法，优化满足吸种条件和效率的最佳播种轨迹，实现按"穴粒数"播补种思想。研究方法将开辟精密播种新技术，提升播种精度水平，具有重要的科学意义和应用前景。

项目组针对超级杂交稻2～3粒/穴，且单粒率少、无空穴的秧盘育秧精密播种目标，研究开发出一种基于机器视觉的智能播种新技术。.首先构建了机器视觉检测系统，研究种子图像特征，统计种子的图像特征与外部形态之间的变化规律，分别建立了灰度值、BP神经网络和改进形状因子的种子图像特征模型库，检测准确率分别达到75%，96.35%和98%。研究了基于BP神经网络的动态毯状秧盘数字化种子群体图像特征检测方法，单连通域内种子数量平均识别准确率率为94.4%，每幅图像平均处理时间0.823s；研究了基于改进形状因子的动态钵体秧盘数字化种子群体图像特征检测方法，穴粒数的平均检测准确率均达到95%以上，每幅图像平均处理时间为0.518s；有效解决了种子粘连、重叠状态下识别准确率低的问题。.设计了带通针的曲柄摆杆伞式吸针，实现了吸针负压吸种、正压通孔过程，解决了吸孔堵塞问题；进行了吸种运动特性分析，确定了吸种的边界条件，即最大速度和加速度分别为1332mm/s、5.08m/s2。创新性提出了“按行蛇形”排布方式划分1×6小区的空穴和单粒穴播种方案，提高播种均匀性，符合移栽机工艺要求。.设计并研制了点位控制播种机构和充填机构，对比分析了最近邻点策略、最短链接策略和穷举法的播种轨迹优化算法，机构运动偏差小于3mm，程序反应时间小于0.61s，且穷举法的程序反应时间仅在0.3s以内，消除了空穴，单粒率小于5%，2粒/穴及以上的合格率达96%，达到了补种精量和投种位置精确的目的。.以单目摄像机作为眼睛，视觉检测系统作为大脑，点播器执行手的播种动作，按照优化的播种轨迹精准控制多个伞式吸针，可以完成多粒、分列、多排的协调作业播种过程，实现了按“穴粒数”播补种的思想，并成功研制一套智能点播机。试验结果表明，当补种率为2%时，单个点播器生产率为215盘/h，双点播器能满足生产率450盘/h的生产需求。.综上所述，项目组完成了研究目标，主要成果有首次应用机器视觉技术对超级杂交稻“毯状秧盘和钵体秧盘”进行了基于BP神经网络和改进形状因子的播种质量在线动态检测方法研究，并创造了一种基于机器视觉的智能播种技术。该方法开辟了智能播种技术领域，提升播种精度水平，研究成果具有重要的科学意义和应用前景。发表论文7篇，EI收录5篇，申请发明专利5项，获得发明专利1项，获得实用新型专利3项，培养青年科技骨干教师4名，研究生4名。