基于移动端轻量模型的杂草分类方法研究

doi:10.13733/j.jcam.issn.20955553.2022.02.023

摘要/Abstract

摘要： 杂草分类的研究大多集中在服务器端模型，而该类模型存在规模较大、占用较多计算资源、计算速度较慢、准确率不高等问题，以ResNet50和MobileNetV3_large模型为基础，采用数据增强、迁移学习来进行快速训练，在训练中通过设置加权Softmax损失函数的权重，最后再利用精度高的服务器端模型指导和优化移动端模型，从而得到一个轻量模型。试验结果表明，本文轻量模型相比移动端模型MobileNetV3_large，在模型大小变化不大的情况下，识别准确率提升1.2%；相比服务器端模型ResNet50，准确率提升0.78%，平均每张推理时间减少7.8%，模型大小减少80%，本研究可为杂草精准施药的实施应用提供理论基础和技术支持。

关键词: 人工智能, 杂草识别, 轻量卷积, 损失函数, 知识蒸馏

Abstract: The research on weed classification mostly focuses on the serverside model, which has the problems of large scale, occupying more computing resources, slow computing speed, and low accuracy. Based on ResNet50 and MobileNetV3_large models, data enhancement and migration learning are used for rapid training. In training, the weight of the weighted Softmax loss function is set. The highprecision serverside model is used to guide and optimize the mobile side model to obtain a lightweight model. The experimental results show that compared with the mobile terminal model MobileNetV3_large, the recognition accuracy of the lightweight model in this paper is improved by 1.2% when the model size changes little. Compared with the serverside model ResNet50, the accuracy is improved by 0.78%, the average reasoning time per piece is reduced by 7.8%, and the model size is reduced by 80%. This study can provide a theoretical basis and technical support for the implementation and application of precision weed application.

Key words: artificial intelligence, weed identification, lightweight convolution, loss function, knowledge distillation

中图分类号:

S126

陈启, 陈慈发, 邓向武, 袁单飞. 基于移动端轻量模型的杂草分类方法研究[J]. 中国农机化学报, 2022, 43(2): 163-170.

Chen Qi, Chen Cifa, Deng Xiangwu, Yuan Danfei.. Research on weed classification method based on mobile lightweight model[J]. Journal of Chinese Agricultural Mechanization, 2022, 43(2): 163-170.

参考文献

［1］　Moazzam S I, Khan U S, Tiwana M I, et al. A review of application of deep learning for weeds and crops classification in agriculture ［C］. 2019 International Conference on Robotics and Automation in Industry (ICRAI), 2019.
［2］　Hamuda E, Glavin M, Jones E. A survey of image processing techniques for plant extraction and segmentation in the field ［J］. Computers and Electronics in Agriculture, 2016, 125: 184-199.
［3］　何东健, 乔永亮, 李攀, 等. 基于SVMDS多特征融合的杂草识别［J］. 农业机械学报, 2013, 44(2): 182-187.
He Dongjian, Qiao Yongliang, Li Pan, et al. Weed recognition based on SVMDS multifeature fusion ［J］. Transactions of the Chinese Society for Agricultural Machinery, 2013, 44(2): 182-187.
［4］　王璨, 李志伟. 利用融合高度与单目图像特征的支持向量机模型识别杂草［J］. 农业工程学报, 2016, 32(15): 165-174.
Wang Can, Li Zhiwei. Weed recognition using SVM model with fusion height and monocular image features ［J］. Transactions of the Chinese Society of Agricultural Engineering, 2016, 32(15): 165-174.
［5］　Fawakherji M, Youssef A, Bloisi D D，et al. Crop and weeds classification for precision agriculture using contextindependent pixelwise segmentation ［C］. IEEE IRC 2019 The Third IEEE International Conference on Robotic Computing. IEEE, 2019.
［6］　Knoll F J, Czymmek V, Poczihoski S, et al. Improving efficiency of organic farming by using a deep learning classification approach ［J］. Journal of Computers and Electronics in Agriculture, 2018, 153: 346-356.
［7］　McCool C, Perez T, Upcroft B. Mixtures of lightweight deep convolutional neural networks: Applied to agricultural robotics ［J］. IEEE Robotics & Automation Letters, 2017: 1344-1351.
［8］　Tang J, Wang D, Zhang Z, et al. Weed identification based on Kmeans feature learning combined with convolutional neural network ［J］. Journal of Computers and Electronics in Agriculture, 2017, 135: 63-70.
［9］　Milioto A, Lottes P, Stachniss C. Realtime blobwise sugar beets vs weeds classification for monitoring fields using convolutional neural networks ［J］. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, 2017.
［10］　Andrea C C, Daniel B B M, Misael J B J. Precise weed and maize classification through convolutional neuronal networks ［C］. 2017 IEEE Second Ecuador Technical Chapters Meeting (ETCM). IEEE, 2017.
［11］　Chavan T R, Nandedkar A V. AgroAVNET for crops and weeds classifification: A step forward in automatic farming ［J］. Computers and Electronics in Agriculture， 2018, 154: 361-372.
［12］　Razavi S, Yalcin H. Using convolutional neural networks for plant classification ［C］. 2017 25th Signal Processing and Communications Applications Conference (SIU), 2017.
［13］　Olsen A, Konovalov D A, Philippa B, et al. DeepWeeds: A multiclass weed species image dataset for deep learning ［J］. Scientific Reports, 2019, 9(1).
［14］　傅雷扬, 李绍稳, 张乐, 等. 田间除草机器人研究进展综述［J］. 机器人, 2021, 43(6): 751-768.
Fu Leiyang, Li Shaowen, Zhang Le, et al. Research progress on field weeding robots: A review ［J］. Robot, 2021, 43(6): 751-768.
［15］　阳湘林. 中国植保无人机产业发展探析［J］. 中国植保导刊, 2020, 40(5): 67-71, 78.
Yang Xianglin. Analysis on the development status of plant protection unmanned aerial vehicle in China ［J］. China Plant Protection, 2020, 40(5): 67-71, 78.
［16］　 He K, Zhang X, Ren S, et al. Deep residual learning for image recognition ［C］. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016.
［17］　Howard A, Sandler M, Chen B, et al. Searching for MobileNetV3 ［C］. 2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019.
［18］　何雪英, 韩忠义, 魏本征. 基于深度卷积神经网络的色素性皮肤病识别分类［J］. 计算机应用, 2018, 38(11): 3236-3240.
He Xueying, Han Zhongyi, Wei Benzheng. Pigmented skin lesion recognition and classification based on deep convolutional neural network ［J］. Journal of Computer Applications, 2018, 38(11): 3236-3240.
［19］　Bagherinezhad H, Horton M, Rastegari M, et al. Label Refinery: Improving ImageNet classification through label progression ［J］. arXiv preprint arXiv: 1805.02641, 2018.
［20］　Yalniz I Z, Jégou H, Chen K, et al. Billionscale semisupervised learning for image classification ［J］. arXiv preprint arXiv: 1905.00546, 2019.
［21］　凌弘毅. 基于知识蒸馏方法的行人属性识别研究［J］. 计算机应用与软件, 2018, 35(10): 181-184, 193.
Ling Hongyi. Pedestrian attribute recognition based on knowledge distillation ［J］. Computer Applications and Software, 2018, 35(10): 181-184, 193.
［22］　高钦泉, 赵岩, 李根, 等. 基于知识蒸馏的超分辨率卷积神经网络压缩方法［J］. 计算机应用, 2019, 39(10): 2802-2808.
Gao Qinquan, Zhao Yan, Li Gen, et al. Compression method of superresolution convolutional neural network based on knowledge distillation ［J］. Journal of Computer Applications, 2019, 39(10): 2802-2808.

[1]	李东升, 胡文泽, 兰玉彬, 范明洪, 李翠云, 朱俊科. 深度学习在杂草识别领域的研究现状与展望[J]. 中国农机化学报, 2022, 43(9): 137-144.
[2]	尚文卿, 齐红波. 基于改进Faster R-CNN与迁移学习的农田杂草识别算法[J]. 中国农机化学报, 2022, 43(10): 176-182.
[3]	侯雨;曹丽英;丁小奇;李静;. 基于边缘检测和BP神经网络的大豆杂草识别研究[J]. 中国农机化学报, 2020, 41(7): 185-190.
[4]	任全会;杨保海;. 图像处理技术在田间杂草识别中应用研究[J]. 中国农机化学报, 2020, 41(6): 154-158.
[5]	杨坪锟;李宇豪;李肖汉;邢彩燕;张晨;霍迎秋;. 基于嵌入式GPU架构的并行形状特征提取算法优化[J]. 中国农机化学报, 2019, 40(2): 193-199.
[6]	李春明;逯杉婷;远松灵;王震洲;. 基于Faster R-CNN的除草机器人杂草识别算法[J]. 中国农机化学报, 2019, 40(12): 171-176.