姜红花,张传银,张昭,毛文华,王东,王东伟.基于Mask R-CNN的玉米田间杂草检测方法[J].农业机械学报,2020,51(6):220-228,247.
JIANG Honghua,ZHANG Chuanyin,ZHANG Zhao,MAO Wenhua,WANG Dong,WANG Dongwei.Detection Method of Corn Weed Based on Mask R-CNN[J].Transactions of the Chinese Society for Agricultural Machinery,2020,51(6):220-228,247.
|
摘要点击次数: 843
全文下载次数: 513
|
| 基于Mask R-CNN的玉米田间杂草检测方法 [下载全文] |
| Detection Method of Corn Weed Based on Mask R-CNN [Download Pdf][in English] |
| 投稿时间：2019-10-08 |
| DOI：10.6041/j.issn.1000-1298.2020.06.023 |
| 中文关键词: 杂草 变量喷药 特征提取 图像分割 残差神经网络 |
| 基金项目:国家重点研发计划项目（2017YFD0700500）、山东省重大科技创新工程项目（2019JZZY010716）、山东省农业重大应用技术创新项目（SD2019NJ001）、山东省重点研发计划项目（2015GNC112004）和山东省自然科学基金项目（ZR2018MC017） |
|
| 中文摘要:针对田间复杂环境下杂草分割精度低的问题，提出了基于Mask R-CNN的杂草检测方法。该方法采用残差神经网络ResNet 101提取涵盖杂草语义、空间信息的特征图；采用区域建议网络对特征图进行杂草与背景的初步二分类、预选框回归训练，利用非极大值抑制算法筛选出感兴趣区域；采用区域特征聚集方法（RoIAlign），取消量化操作带来的边框位置偏差，并将感兴趣区域（RoI）特征图转换为固定尺寸的特征图；输出模块针对每个RoI计算分类、回归、分割损失，通过训练预测候选区域的类别、位置、轮廓，实现杂草检测及轮廓分割。在玉米、杂草数据集上进行测试，当交并比（IoU）为0.5时，本文方法均值平均精度 （mAP）为0.853，优于SharpMask、DeepMask的0.816、0.795，本文方法的单样本耗时为280ms，说明本文方法可快速、准确检测分割出杂草类别、位置和轮廓，优于SharpMask、DeepMask实例分割算法。在复杂背景下对玉米、杂草图像进行测试，在IoU为0.5时，本文方法mAP为0.785，单样本耗时为285ms，说明本文方法可实现复杂背景下的农田作物杂草分割。在田间变量喷洒试验中，杂草识别准确率为91%，识别出杂草并准确喷雾的准确率为85%，准确喷药的杂草雾滴覆盖密度为55个/cm2，装置对每幅图像的平均处理时间为0.98s，满足农药变量喷洒的控制要求。 |
| JIANG Honghua ZHANG Chuanyin ZHANG Zhao MAO Wenhua WANG Dong WANG Dongwei |
| Shandong Agricultural University;Baoji University of Arts and Sciences;Chinese Academy of Agricultural Mechanization Sciences; Qingdao Agricultural University |
| Key Words:weeds variable spraying pesticide feature extraction image segmentation residual neural network |
| Abstract:Accurate detection and identification of weeds is a prerequisite for weed control. Aiming at the problem of low accuracy of weed segmentation in complex field environment, an intelligent weed detection and segmentation method based on Mask R CNN was proposed. The ResNet 101 network was used to extract the feature map of weed semantic and spatial information. The characteristic map was classified by the regional suggestion network, and the pre selection box regression was trained. The pre selection area was screened by the non maximum suppression algorithm. RoIAlign was used to cancel the border position deviation caused by quantization, and the region of interest (RoI) feature map was transformed into a fixed size feature map. The output module calculated the classification, regression and segmentation loss for each RoI, predicted the category, location and contour of the candidate area through training, and realized weed detection and contour segmentation. When IoU (intersection over union) was 0.5, the mean accuracy precision (mAP) value was 0853, which was better than that of SharpMask and DeepMask with 0.816 and 0.795, respectively. The single sample time of the three methods was 280ms, 256ms and 248ms respectively. The results showed that the method can quickly and accurately detect and segment the category, location and contour of weeds, and it can be better than SharpMask and DeepMask. When IoU was 0.5, the mAP value of the proposed method was 0.785, and the time for a single sample was 285ms, indicating that this method can realize the field operation in the complex background and meet the real time control requirements of field pesticide variable spraying. In the field variable spraying test, the accuracy rate of identifying weeds was 91%, the accuracy rate of identifying weeds and spraying them accurately was 85%, the spray density of pesticide spray droplets was 55 per square centimetre, and the average processing time of the device was 0.98s. It can meet the control standard of pesticide variable spraying. |
Transactions of the Chinese Society for Agriculture Machinery (CSAM), in charged of China Association for Science and Technology (CAST), sponsored by CSAM and Chinese Academy of Agricultural Mechanization Science(CAAMS), started publication in 1957. It is the earliest interdisciplinary journal in Chinese which combines agricultural and engineering. It always closely grasps the development direction of agriculture engineering disciplines and the published papers represent the highest academic level of agriculture engineering in China. Currently, nearly 8,000 papers have been already published. There are around 3,000 papers contributed to the journal each year, but only around 600 of them will be accepted. Transactions of CSAM focuses on a wide range of agricultural machinery, irrigation, electronics, robotics, agro-products engineering, biological energy, agricultural structures and environment and more. Subjects in Transactions of the CSAM have been embodied by many internationally well-known index systems, such as: EI Compendex, CA, CSA, etc.
|
| 下载PDF阅读器 |
|
|
|
