In order to significantly enhances operational efficiency and provides a highly feasible technical solution for the upgrading of modern agriculture, this research is dedicated to the design and development of an automatic citrus fruits picking system based on machine vision. The primary objective is to achieve automated localization and recognition of citrus fruits, thereby offering robust technical support for unmanned orchard operations. The study leverages a drone platform to capture two-dimensional image data of citrus fruits and constructs an advanced image processing algorithm pipeline based on computer vision. The experimental results reveal that the automatic picking system maintains high accuracy in citrus fruits recognition across varying lighting conditions and complex backgrounds, demonstrating significant robustness.