基于FCN网格定位和特征融合的水培生菜幼苗状态检测方法

为了及时发现问题幼苗状态和提高幼苗分拣效率，本文以水培生菜幼苗培育过程中出现的死亡和双株状态为研究对象，提出一种基于FCN网格定位和特征融合的水培生菜幼苗状态检测方法。本方法在原有研究的基础上，针对双株状态幼苗检测精度低的问题，引入FCN架构改变原有边框回归方式，利用其对位置信息敏感的特性，获取精确的网格点空间信息。同时，利用特征融合策略，充分获取不同网格点间的相关性，实现对水培生菜幼苗问题状态的精准定位。实验结果表明，该方法的平均检测精度达到了88.1%，检测精度优于原有方法、FSAF、YOLO V3、FoveaBox、ATSS和CornerNet，尤其对双株状态的幼苗检测精度得到明显提升。因此，本文提出的方法能够实现水培生菜幼苗问题状态的自动检测，为水培蔬菜育苗分拣智能化及种植自动化提供技术支持。

A Detection Method of Hydroponic Lettuce Seedlings Status Based on FCN Grid Location and Feature Fusion

In order to find problematic seedlings status timely and improve sorting efficiency of seedlings in the cultivation stage of hydroponic vegetables, this paper proposed an automatic detection method of hydroponic lettuce seedlings status based on FCN grid location and feature fusion, taking the dead and double-planting status of seedlings growing in a hole as our research object. Aiming at the problem of low detection accuracy of two-plant seedlings status, our approach introduced FCN architecture to change the traditional localization based on regression and adopted its sensitive characteristics to obtain accurate grid point spatial information on the basis of the previous research. At the same time, the feature fusion strategy was used to fully obtain the correlation between different grid points, so as to achieve further accurate location of the problematic status of hydroponic lettuce seedlings. The experimental results showed that the mean average precision of this method was 88.1%, which was higher than that of the previous method, FSAF, YOLO V3, FoveaBox, ATSS and CornerNet. In particular, the detection accuracy of two-plant seedlings status was significantly increased. Therefore, the hydroponic lettuce seedling condition detection method proposed in this paper can realize the automatic detection of the problem status of hydroponic lettuce seedlings, and provide technical support for the intelligent breeding and sorting of hydroponic vegetable seedlings and planting automation. Therefore, the method proposed in this paper can realize accurate identification and localization automatically, which can provide technical support for intelligent sorting and automatic planting of hydroponic vegetable seedlings.