As a characteristic agricultural product of Fengjie, Chongqing, navel orange picking faces inefficient manual picking and labor shortages, thus it is very important to promote their automatic picking. To enhance the automation of the navel orange picking robot, this article proposes a novel navel orange recognition algorithm that addresses the challenges of light changes, stacked fruits, and foliage occlusion in complex environments, as well as the excessive parameters and computational complexity of existing recognition models. The algorithm first divides the image into grids, then randomly selects points from the grids to draw circles, and eliminates contour points passed by circles to reduce the number of circles, which can improve the speed of navel orange recognition. To optimize the recognition results, the algorithm removes low-coverage circles and overlapping circles to retain circles that better fit the navel oranges, and eliminates noisy circles to avoid interference from distantly located navel oranges. Using 417 navel orange images collected from a navel orange planting base in Chongqing as the dataset, the algorithm achieves a precision rate of 96.5%, a recall rate of 94.3%, and an F1 score of 0.954, which demonstrates that the algorithm provides efficient and precise fruit recognition capabilities for navel orange automated picking robot.