基于轻量化 Transformer的农作物检测机器人路径规划

doi:10.13733/j.jcam.issn.2095-5553.2024.09.035

摘要/Abstract

摘要：

为解决农作物检测机器人路径规划算法在复杂农田环境下精度低、速度慢等问题，设计轻量化 Transformer模型，将其用于农作物检测机器人的路径规划任务中。采用余弦函数替代 softmax计算，使得查询、键、值向量的计算可拆分，时间复杂度由原来的 Ο（N 2）降低到 Ο（N）。通过四种不同聚合方式处理特征向量，确定节点权重分配。试验结果表明，基于轻量化 Transformer的农作物检测机器人路径规划方法能够显著提高农作物检测机器人的效率和准确性。相比传统的规则化路径规划算法，将 100规模农作物检测机器人的路径长度缩短 5. 91%；相比 Transformer模型，推理时间缩短 50%，训练时间缩短 75%，为农作物检测机器人的路径规划提供一种新颖且有效的解决方案。

关键词: 农作物检测机器人, 轻量化 Transformer, 强化学习, 多智能体, 路径规划

Abstract:

In order to solve the problems such as low precision and slow speed of path planning algorithm of crop inspection robot in complex farmland environment，this paper designed a lightweight Transformer model and applied it to the path planning task of crop inspection robots. The cosine function was used to replace the softmax calculation，so that the computation of query，key and value vectors could be split，and the time complexity was reduced from Ο（N 2）to Ο（N）. Four different aggregation methods were used to process the feature vectors and determine the node weight allocation. The experimental results showed that the path planning method of crop detection robot based on lightweight Transformer could significantly improve the efficiency and accuracy of crop detection robot. Compared with the traditional regularized path planning algorithm，it shortened the path length of 100.scale crop detection robot by 5. 91%. Compared with the Transformer model，it reduced the inference time by 50% and the training time by 75%. It provided a novel and effective solution for the path planning of crop detection robot.

Key words:

"> crop detection robot, lightweight Transformer, reinforcement learning, multi agent, path planning

中图分类号:

TP242
S23

李娟 , 金志雄 . 基于轻量化 Transformer的农作物检测机器人路径规划[J]. 中国农机化学报, 2024, 45(9): 227-233.

Li Juan, Jin Zhixiong. Path planning of crop inspection robot based on lightweight Transformer [J]. Journal of Chinese Agricultural Mechanization, 2024, 45(9): 227-233.

参考文献

［ 1］李瑞，敖雁，孙启洵，等 .大田农业物联网应用现状与展望［ J］.北方园艺， 2018（14）：148-153.

Li Rui， Ao Yan， Sun Qixun， et al. Application and prosprct of agricultural interent of things in crop field.［J］. Northern Horticulture，2018（14）：148-153.

［ 2］李晓静，余东满 .改进蚁群算法在果蔬采摘机器人路径规划中的应用［ J］.江苏农业科学， 2018，46（23）：253-258.

Li Xiaojing，Yu Dongman. Application of improved ant colony algorithm in path planning of fruit and vegetable picking robot［J］. Jiangsu Agricultural Sciences， 2018，46（23）： 253-258.

［ 3］崔永杰，王寅初，何智，等 .基于改进 RRT算法的猕猴桃采摘机器人全局路径规划［ J］.农业机械学报， 2022，53（6）：151-158.

Cui Yongjie，Wang Yinchu，He Zhi，et al. Global path planning of kiwifruit harvesting robot based on improved RRT algorithm［J］. Transactions of the Chinese Society for Agricultural Machinery，2022，53（6）：151-158.

［ 4］刘顺，闫荣兰，台浩娇，等 .基于局部势场 A*算法和动态窗口法的农用轮式机器人路径规划［ J］.江苏农业科学， 2022，50（2）：192-198.

Liu Shun，Yan Ronglan，Tai Haojiao，et al. Path planning of agricultural wheeled robot based on local potential field A* algorithm and dynamic window method［J］. Jiangsu Agricultural Sciences，2022，50（2）：192-198.

［ 5］陈娟，姚立健，徐丽君，等 .基于无模型自适应控制的设施园艺自动导航车辆路径跟踪［ J］.江苏农业科学， 2022，50（8）：187-192.

Chen Juan，Yao Lijian，Xu Lijun，et al. Path tracking of automatic navigation vehicle in protected horticulture based on model.free adaptive controller［J］. Jiangsu Agricultural Sciences，2022，50（8）：187 -192.

［ 6］王辉，王景良，朱龙彪，等 .多功能农用机器人路径规划的多种优化算法［ J］.江苏农业科学， 2018，46（5）： 199-203.

Wang Hui，Wang Jingliang，Zhu Longbiao，et al. Path planning of multi.functional agricultural robot based on multiple optimization algorithms［J］. Jiangsu Agricultural Sciences，2018，46（5）：199-203.

［ 7］Bengio Y， Lodi A， Prouvost A. Machine learning for combinatorial optimization： A methodological tour d′horizon［J］. European Journal of Operational Research， 2021，290（2）：405-421.

［ 8］王原，陈名，邢立宁，等 .用于求解旅行商问题的深度智慧型蚁群优化算法［ J］.计算机研究与发展， 2021，58 （8）：1586-1598.

Wang Yuan， Chen Ming， Xing Li.Ning， et al. Deep intelligent ant colony optimization algorithm for solving traveling salesman problem［J］. Journal of Computer Research and Development，2021，58（8）：1586-1598.

［ 9］Xin L，Song W，Cao Z，et al. Multi.decoder attention model with embedding glimpse for solving vehicle routing problems［C］. Proceedings of the AAAI Conference on Artificial Intelligence，2021，35（13）：12042-12049.

［10］Vaswani A，Shazeer N，Parmar N，et al. Attention is all you need［C］. Proceedings of the 31st International Conference on Neural Information Processing Systems， 2017：6000-6010.

［11］王红君，付勇，岳有军，等 .基于并行蚁群算法的设施温室机器人多点路径规划的研究［ J］.江苏农业科学， 2019，47（17）：237-241.

Wang Hongjun，Fu Yong，Yue Youjun，et al. Study on multi.point path planning of facility greenhouse robot based on parallel ant colony algorithm［J］. Jiangsu Agricultural Sciences，2019，47（17）：237-241.

［12］Vinyals O，Fortunato M，Jaitly N. Pointer networks［J］. arXiv：1506. 03134，2015.

［13］Nazari M，Oroojlooy A， Snyder L，et al. Reinforcement learning for solving the vehicle routing problem［C］. Proceedings of the 32st International Conference on Neural Information Processing Systems，2018：9839-9849.

［14］Kool W，Van hoof H，Welling M. Attention，learn to solve routing problems［C］. Proceedings of the International Conference on Learning Representations，2019.

［15］Li J W，Xin L，Cao Z G，et al. Heterogeneous attentions for solving pickup and delivery problem via deep reinforcement learning［J］. IEEE Transactions on Intelligent Transportation Systems，2021，23（3）：2306-2315.

［16］Williams R J. Simple statistical gradient.following algorithms for connectionist reinforcement learning［J］. Machine Learning，1992，8（3-4）：229-256.

［17］王扬，陈智斌，杨笑笑，等 .深度强化学习结合图注意力模型求解 TSP问题［ J］.南京大学学报（自然科学版）， 2022，58（3）：420-429.

Wang Yang，Chen Zhibin，Yang Xiaoxiao，et al. Deep reinforcement learning combined with graph attention model for solving TSP［J］. Journal of Nanjing University（Natural Science），2022，58（3）：420-429.

［18］Nabaei A，Hamian M，Parsaei M R，et al. Topologies and performance of intelligent algorithms： A comprehensive review［J］. Artificial Intelligence Review，2018，49（1）： 79-103.