Abstract: Agricultural machinery vision navigation plays a crucial role in residual film recycling and interrow cultivation. However, due to the complex background of actual farmland images, it is difficult for traditional algorithms to extract navigation lines effectively. To solve this problem, this paper employs the LinkNet model to complete the realtime segmentation of the target areas and backgrounds in farmland images. In terms of improving the accuracy of navigation line fitting, a navigation line extraction model is designed based on the orthogonal regression equation, and a navigation line fitting method using the chaotic Lévy flight cloning artificial fish school optimization algorithm （CLCAFSA） is proposed. The CLCAFSA is experimentally compared with improved algorithms based on the sparrow search algorithm （SSA）, circle search algorithm （CSA）, and particle swarm optimization （PSO） in terms of fitting accuracy. The experimental results show that CLCAFSA effectively solves the problem of poor segmentation quality in farmland images and achieves accurate navigation line fitting. The orthogonal distance and average deviation of the CLCAFSA algorithm are reduced by 12.96%, 10.44% and 6.55% on average compared to CSSA, CCSA and CPSO, respectively, which significantly improves the fitting accuracy of navigation lines.

Key words: agricultural machinery vision navigation, swarm intelligence, navigation line fitting, least squares method, artificial fish swarm optimization algorithm, semantic segmentation

摘要： 农机视觉导航在残膜回收和中耕作业中具有重要意义，但由于实际农田图像背景复杂，传统算法难以完成导航线提取。为解决这一难题，通过LinkNet模型完成农田图像的目标区域与背景实时分割。在提高导航线拟合精度方面，基于正交回归方程设计导航线提取模型，提出混沌莱维飞行克隆人工鱼群优化算法（CLCAFSA）的导航线拟合方法获取导航线。将CLCAFSA与基于麻雀算法（SSA）、圆圈搜索算法（CSA）和粒子群算法（PSO）的改进算法在拟合精度上进行试验对比。结果表明，CLCAFSA能有效解决农田图像分割效果不理想状态下导航线精确拟合问题。CLCAFSA优化后的正交距离和比CSSA、CCSA和CPSO分别降低12.96%、10.44%和6.55%，显著提高导航线拟合精度。

关键词: 农机视觉导航, 群智能, 导航线拟合, 最小二乘法, 人工鱼群优化算法, 语义分割