基于RGE—UNet模型的甘蔗蔗梢识别研究

doi:10.13733/j.jcam.issn.2095-5553.2025.03.028

摘要/Abstract

摘要： 传统的甘蔗蔗梢图像分割算法步骤烦琐、整体优化较为困难，采用在小样本上仍表现优异的UNet网络，将模型原有主干网络替换为ResNet50来简化模型训练过程，上采样部分用Ghost轻量级模块替换普通卷积模块以减少模型的参数量和浮点数计算量，同时在编码器和解码器之间加入SE注意力机制对提取到的特征权重进行优化，最终得到一个轻量级的RGE—UNet蔗梢分割模型。结合迁移学习的方法对模型进行训练，将训练完成的模型通过Canny算子与水平垂直投影法对蔗梢区域进行识别，并提取蔗梢分割路径坐标。结果表明，基于RGE—UNet模型识别方法的平均像素准确率为94.98%，单张图片分割时间为0.31 s，分割速度较UNet与R50—UNet模型分别提高13.9%和18.4%。该模型能实现对蔗梢的快速准确识别，为甘蔗收割的自动化研究提供一定的技术参考。

关键词: 甘蔗蔗梢, 路径识别, 语义分割, RGE—UNet, 迁移学习

Abstract:

In response to the problems of cumbersome steps and difficulty in overall optimization of traditional image segmentation algorithms for sugarcane shoots, this paper adopts a UNet network, which performs well on small samples. The original backbone network of the model is replaced with ResNet50 to simplify the model training process. In the upsampling part, the Ghost lightweight module is used to replace the ordinary convolution module to reduce the number of model parameters and floating-point operations. An SE attention mechanism is added between the encoder and decoder to optimize the extracted feature weights. The result is a lightweight RGE—UNet sugarcane shoot segmentation model. Then, the trained model identifies the shoot region using the Canny operator and the horizontal/vertical projection methods and extracts the segmentation path coordinates. The experimental results show that the recognition method based on the RGE—UNet model achieves an average pixel accuracy of 94.98%, a segmentation time of 0.31 seconds per image, improving segmentation speed by 13.9% and 18.4% compared with the UNet and R50—UNet models, respectively. This model achieves fast and accurate recognition of sugarcane shoots and provides technical references for the automation of sugarcane harvesting.

Key words: , sugarcane shoots, path recognition, semantic segmentation, RGE—UNet, transfer learning

中图分类号:

沈中华, 程虎强, , 夏爱强, 李涵. 基于RGE—UNet模型的甘蔗蔗梢识别研究[J]. 中国农机化学报, 2025, 46(3): 188-194.

Shen Zhonghua, Cheng Huqiang, , Xia Aiqiang, Li Han. Research on sugarcane shoot recognition based on RGE—UNet model[J]. Journal of Chinese Agricultural Mechanization, 2025, 46(3): 188-194.

参考文献

［1］　肖威, 陆静平. 甘蔗机械化收获技术现状分析［J］. 中国农机化学报, 2022, 43（2）: 50-59, 142.
Xiao Wei, Lu Jingping. Analysis of sugarcane mechanized harvesting technology ［J］. Journal of Chinese Agricultural Mechanization, 2022,43 （2）: 50-59, 142.
［2］　罗菊川, 区颖刚, 刘庆庭, 等. 整秆式甘蔗联合收获机断蔗尾机构［J］. 农业机械学报, 2013, 44（4）: 89-94, 107.
Luo Juchuan, Qu Yinggang, Liu Qingting, et al. Tailbreaking mechanism of whole stalk sugarcane combine harvester ［J］. Transactions of the Chinese Society for Agricultural Machinery, 2013, 44（4）: 89-94, 107.
［3］　Xie L, Wang J, Cheng S, et al. Optimisation and dynamic simulation of a conveying and top breaking system for wholestalk sugarcane harvesters ［J］. Biosystems Engineering, 2020, 197: 156-169.
［4］　谢卢鑫, 王俊, 程绍明, 等. 甘蔗收获机剥叶断尾系统的设计与转速优化［J］. 农业工程学报, 2016, 32（6）: 50-57.
Xie Luxin, Wang Jun, Cheng Shaoming, et al. Design and speed optimization of leafcrushing and tailbreaking system of sugarcane harvester ［J］. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32（6）: 50-57.
［5］　沈中华, 李尚平, 麻芳兰, 等. 小型甘蔗收获机喂入能力仿真与试验［J］. 农业机械学报, 2014, 45（11）: 117-123.
Shen Zhonghua, Li Shangping, Ma Fanglan, et al. Simulation and experiment on feed ability of small sugarcane harvester ［J］. Transactions of the Chinese Society for Agricultural Machinery, 2014, 45 （11）: 117-123.
［6］　麻芳兰, 蒋红梅, 李尚平, 等. 整秆式甘蔗收获机剥叶断尾机构设计与试验［J］. 农业机械学报, 2012, 43（6）: 73-78,7.〖JP2〗Ma Fanglan, Jiang Hongmei, Li Shangping, et al. Design and experiment on cleaning leaves and breaking tails mechanism of whole stalk sugarcane harvester ［J］.〖JP〗 Transactions of the Chinese Society for Agricultural Machinery, 2012, 43（6）: 73-78, 7.
［7］　杨涛, 李晓晓. 机器视觉技术在现代农业生产中的研究进展［J］. 中国农机化学报, 2021, 42（3）: 171-181.
Yang Tao, Li Xiaoxiao. Research progress of machine vision technology in modern agricultural production ［J］. Journal of Chinese Agricultural Mechanization, 2021, 42（3）: 171-181.
［8］　Liu G X, Mao S Y, Kim J H. A maturetomato detection algorithm using machine learning and color analysis ［J］. Sensors, 2019, 19(9): 2023.
［9］　Kadir M F A, Yusri N A N, Rizon M, et al. Automatic mango detection using texture analysis and randomised hough transform ［J］. Applied Mathematical Sciences, 2015, 9（129）: 6427-6436.

［10］　张小花, 马瑞峻, 吴卓葵, 等. 基于机器视觉的果园成熟柑橘快速识别及产量预估研究［J］. 广东农业科学, 2019, 46（7）: 156-161.

Zhang Xiaohua, Ma Ruijun, Wu Zhuokui, et al. Fast detection and yield estimation of ripe citrus fruit based on machine vision ［J］. Guangdong Agricultural Sciences, 2019, 46（7）: 156-161.

［11］　赵博杰. 基于机器视觉的茶叶嫩芽识别关键技术研究［D］. 鞍山:辽宁科技大学, 2020.

Zhao Bojie. Tea bud segmentation and recognition based on machine vision ［D］. Anshan: University of Science and Technology Liaoning, 2020.

［12］　韩振浩, 李佳, 苑严伟, 等. 基于U—Net网络的果园视觉导航路径识别方法［J］. 农业机械学报, 2021, 52（1）: 30-39.
Han Zhenhao, Li Jia, Yuan Yanwei, et al. Path recognition of orchard visual navigation based on U—Net ［J］. Transactions of the Chinese Society for Agricultural Machinery, 2021, 52（1）: 30-39.

［13］　张佳彬. 基于深度学习的头足类动物角质颚色素沉积量化方法研究［D］. 上海: 上海海洋大学, 2022.

Zhang Jiabin. A quantitative method of pigmentation in beak of cephalopods based on deep learning ［D］. Shanghai: Shanghai Ocean University, 2022.

［14］　朱成宇. 基于轻量级卷积神经网络的水稻病害识别［D］. 南宁: 广西大学, 2022.Zhu Chengyu. Rice disease identification based on lightweight convolutional neural network ［D］. Nanning: Guangxi University, 2022.
［15］　马冬梅, 杨彩锋, 李鹏辉. 多尺度特征融合的图像语义分割［J］. 西北师范大学学报（自然科学版）, 2020, 56（1）: 44-50.Ma Dongmei, Yang Caifeng, Li Penghui. Image semantic segmentation based on multiscale feature fusion ［J］. Journal of Northwest Normal University （Natural Science）, 2020, 56 （1）: 44-50.

［16］　胡明玉. 基于深度学习的苹果树芽体图像识别方法研究［D］. 北京: 中国农业科学院, 2021.

Hu Mingyu. Research on apple tree bud image recognition method based on deep learning ［D］. Beijing: Chinese Academy of Agricultural Sciences, 2021.

［17］　钟志峰, 何佳伟, 侯瑞洁, 等. 改进UNet的轻量化道路图像语义分割算法［J］. 现代电子技术, 2022, 45（19）: 71-76.

Zhong Zhifeng, He Jiawei, Hou Ruijie, et al. Lightweight road image semantic segmentation algorithm based on improved UNet ［J］. Modern Electronics Technique, 2022, 45 （19）: 71-76.

［18］　徐旭, 陈曦. 基于FPGA的sobel边缘检测算法的实现［J］. 电子测试, 2022, 36（23）: 63-65.

Xu Xu, Chen Xi. Implementation of sobel edge detection algorithm based on FPGA ［J］. Electronic Test, 2022, 36 （23）: 63-65.

［19］　王嘉豪. 基于机器视觉的铝车身缺陷特征在线检测技术研究［D］. 天津: 天津理工大学, 2022.

Wang Jiahao. Research on online defect feature detection technology of aluminum body based on machine vision ［D］. Tianjin: Tianjin University of Technology, 2022.

［20］　刘博一. 基于双目立体视觉的焊缝定位方法研究［D］. 哈尔滨: 哈尔滨工业大学, 2018.

Liu Boyi. Research on weld location method based on binocular stereo vision ［D］. Harbin: Harbin Institute of Technology, 2018.

[1]	王丹阳, 梁伟红, 李玉萍, 黄贵修, . 面向不平衡数据的木薯叶部病害图像识别方法[J]. 中国农机化学报, 2025, 46(3): 101-107.
[2]	刘起, 代东南, 孙孟研, 马德新, 徐阳. 基于双重注意力融合网络结构的圣女果果实表面缺陷识别[J]. 中国农机化学报, 2025, 46(3): 153-159.
[3]	郭敬涛, 吕凤, 章慧婷, 杨彪, 刘大洋. 基于迁移学习与轻量化YOLOv5s的草莓目标检测方法[J]. 中国农机化学报, 2025, 46(3): 253-260.
[4]	曾德斌, 陆万荣, 郑良芳, 施明登, 陈文绪, . 基于U2—DSCNet植物叶片分割方法研究[J]. 中国农机化学报, 2025, 46(2): 253-258.
[5]	段小勇, 何超, 刘学渊. 基于改进DeepLabV3+的非结构化道路可行驶区域检测[J]. 中国农机化学报, 2025, 46(2): 271-278.
[6]	李春雨, 郭肖琴, 杨晶晶, 李忠华. 基于深度学习的水果图像识别[J]. 中国农机化学报, 2025, 46(1): 198-203.
[7]	牛潘婷, 张宝林, 潘丽杰, 郭建鹏(. 基于改进GoogLeNet的玉米叶片病害识别及其可解释性研究[J]. 中国农机化学报, 2025, 46(1): 204-212.
[8]	赵爽, 俞永强, 苗玉彬, 刘可心 . 基于改进 Mask R-CNN的青菜杂质检测研究[J]. 中国农机化学报, 2024, 45(9): 77-82.
[9]	韩鹏飞, 宋其江, 贾梦实. 基于改进轻量化 EfficientNet-V2模型的小麦种子分类[J]. 中国农机化学报, 2024, 45(9): 111-117.
[10]	宋小鹿 , 冯娟 , 梁翔宇 , 周玺兴 . 基于 YOLOv4的水下海参检测与计数算法[J]. 中国农机化学报, 2024, 45(9): 258-264.
[11]	刘星, 顾寄南, 黄则栋, 张文浩, 张伟. 基于深度学习的苹果点云语义分割方法研究[J]. 中国农机化学报, 2024, 45(8): 223-227.
[12]	苏斐, 王光辉, 史艳霞, 贾然, 闫银发, 祖林禄, . 融合通道注意力机制的ResUnet作物病斑分割模型[J]. 中国农机化学报, 2024, 45(8): 228-233.
[13]	郭宏杰, 马睿, 王佳, 赵威, 马德新. 基于卷积神经网络的苹果叶部病害图像识别研究[J]. 中国农机化学报, 2024, 45(5): 239-245.
[14]	王化佳, 顾寄南, 王梦妮, 夏子林. 基于深度学习的茶嫩芽分割与采摘点定位方法研究[J]. 中国农机化学报, 2024, 45(5): 246-252.
[15]	邱金凯, 许秀英, 康烨, 臧浩, 马锴, 郭志鹏. 基于语义分割的无人机图像西瓜果实提取研究[J]. 中国农机化学报, 2024, 45(3): 182-188.