共检索到2448条,权限内显示50条;
[学术文献 ] 基于双重稳健回归的果树行间可行驶区域识别算法 进入全文
期刊：华南农业大学学报
【目的】提出一种复杂环境下以天空为背景的果树行间可行驶区域识别算法，以便农业机器人导航系统中工作路径的提取。【方法】通过蓝色分量(B分量)进行树冠和背景天空的分离，改进Otsu算法实现更好的分割效果，形态学处理后根据树顶分布规律，进行动态阈值“V形”感兴趣区域寻找及特征点提取，使用泰尔-森稳健回归剔除干扰点后，使用随机采样一致性(Random sample consensus,RANSAC)算法进行拟合，得到树顶处直线，通过斜率变换关系得到可行驶区域边缘直线斜率，利用剔除后特征点信息和剔除阈值获得关键点坐标，以斜率为约束条件，代入关键点，得到可行驶区域边缘直线方程，并使用最小二乘法进行拟合，以此实现可行驶区域识别。【结果】试验结果表明，本文双重稳健回归算法较泰尔-森算法和RANSAC算法平均偏差角度分别减小了8.28%和9.88%，标准差分别减少了6.25%和22.89%，准确率分别提高了4.64%和10.49%。【结论】研究结果可为农业机器人在大多数标准化果园复杂环境中的可行驶区域识别和路径提取提供研究思路。
[相关成果 ] A Smart Agriculture Storaging System with IOT Based 进入全文
IEEE
We are designing the smart agriculture storage room. Now a days many machines are available to the farmers for cultivating purpose. Recently technology was developed with the help of smart phones. Farmers are using smart phones. Based on that we have implemented that storage room to control with the help of smart phones. In this research work, a part of smart agriculture storage technology is done by using the Bluetooth in mobile device. This paper describes about storage room which is placed in the agriculture itself. In the storage room different type loads are placed by using the different components like Arduino, 4 Channel relay network and Bluetooth. The farmers can control the storage room with enabled phone with Bluetooth. The farmers can easily access and controlled the storage room by staying at particular place and access them with phone without the help of other people.
[会议论文 ] TinyML Smart Sensor for Energy Saving in Internet of Things Precision Agriculture platform 进入全文
IEEE
Smart agriculture researchers bring numerous tools and prospects to the farm ecosystem to improve its productivity and, mainly, its sustainability. Artificial Intelligence (AI) is widely used in precision agriculture as Internet of Things (IoT) technologies have brought a huge volume of data to exploit to provide useful insights for farmers such as weather prediction, pest development detection, or harvest time estimation. AI algorithms are mostly executed in the cloud due to their inherent computing constraints, thus requiring the different sensors to offload their data to the appropriate server. Depending on the amount and volume of data exchanged, the need for computer offloading may induce privacy, security, and latency issues in addition to weighting on the sensor’s battery consumption as wireless transmission methods have a high-energy demand. To overcome this difficulty, recent research has tried to bring AI computation closer to the end device with edge or fog computing and more recently with the Tiny Machine Learning (TinyML) paradigm that aims to embed the AI algorithm directly into the sensor’s microcontroller. In that context, this paper proposes a prototype of smart sensor capable of detecting fruits presence with TinyML. We then study the energy consumption of our system in different IoT scenarios.
[学术文献 ] 基于密度自适应的RANSAC非结构化环境下果园机器人导航 进入全文
期刊：华南农业大学学报
【目的】提出一种基于多传感器融合的果园导航方案，解决果园机器人在GPS导航过程中受果树遮挡导致信号弱、定位效果差的问题。【方法】通过16线激光雷达采集高精度的三维点云数据，利用Voxel grid filter滤波算法进行点云预处理，降低点云密度并去除离散点，将果树行通过欧几里类算法进行聚类，采用改进的随机采样一致性(Random sample consensus, RANSAC)算法拟合出果树行直线，根据平行直线的关系，推算得到导航线，并融合惯性测量单元(Inertial measurement unit, IMU)对果园机器人进行高精度定位。基于差速转向和纯追踪模型进行轨迹跟踪，实现果园机器人在果树行间自主导航以及自动换行的目标。【结果】在将激光雷达和IMU的数据进行融合后，获取到果园机器人的准确位姿，当机器人以速度0.8 m/s在果园作业时，对比最小二乘法和传统RANSAC法产生的偏差，基于密度自适应RANSAC法产生的横向偏差不超过0.1 m、航向角偏差不超过1.5°，均为3种方法中的最小值。但当机器人速度增加到1.0 m/s时，各项偏差均明显增大。【结论】本文提出的基于多传感器融合的果园机器人导航技术适用于大多数规范化果园，具有重要推广价值。
[学术文献 ] 生态农业发展的回顾与展望 进入全文
期刊：华南农业大学学报
生态农业在我国正值大发展之际，针对目前生态农业一些模糊认识和畏难情绪，文章回顾了生态农业产生的历史轨迹和现实缘由，追溯了国内外对农业可持续发展道路的各类探索，概括了生态农业概念的内涵和理论基础，总结了践行生态农业的多种切入途径。对生态农业成为主流农业方式需要解决的效应判断、潜力挖掘、市场创新、政策支撑和民间参与等进行了讨论，并对生态农业今后蓬勃发展的机遇与态势进行了展望。
[学术文献 ] 农机虚拟装配分类检测网络数据集构建方法 进入全文
期刊：华南农业大学学报
【目的】虚拟装配在工业中可节约生产成本,提升机械部件生产效率。现有的虚拟现实引擎缺乏自动建立碰撞体功能,无法完整还原实际装配过程中的物理属性;通用化构建零件网格实体是提升虚拟装配实用性、精确性及通用性的重要途径。【方法】以批量农机部件为样本对象,设计批量预处理算法及改进采样相关算法,通过构建三维模型样本的图片数据集,训练人工智能分类检测网络,从图片中分类并检测相关参数,实现自动构建碰撞体功能。【结果】经过优化算法处理训练得到的分类检测网络从图片分类零件种类的精度在98%以上,从图片检测零件各项碰撞体构建参数的精度在98%以上;未经优化处理训练的网络不收敛。【结论】本研究方法可以有效提升人工智能分类检测网络的识别精度及训练效率,结合碰撞体参数化构建程序,可提升碰撞体建模精度。
