Improved honeybee pollination lightweight visual detection method

The invention discloses an improved honeybee pollination lightweight visual detection method, which comprises the following steps: acquiring a honeybee pollination image data set, and preprocessing the honeybee pollination image data set; an improved lightweight YOLOv5s convolutional neural network model is built; the improved lightweight YOLOv5s convolutional neural network model is trained; and inputting a to-be-detected bee pollination image into the trained YOLOv5s convolutional neural network model to obtain the type and position information of bee pollination. The method has relatively good stability and robustness, has relatively high detection accuracy on bee pollination pictures in a complex environment background, and can realize accurate detection of bee pollination activities in a complex environment. 本发明公开了一种改进的蜜蜂授粉轻量化视觉检测方法，包括：获取蜜蜂授粉图像数据集，并进行预处理；搭建改进的轻量化YOLOv5s卷积神经网络模型；训练改进后的轻量化YOLOv5s卷积神经网络模型；将待测蜜蜂授粉图像输入至训练后的YOLOv5s卷积神经网络模型，得到蜜蜂授粉的类别和位置信息。本方法具备较好的稳定性和鲁棒性，对复杂环境背景下的蜜蜂授粉图片有较高的检测准确率，可实现复杂环境下精确检测蜜蜂授