Polyphyletic Loss: Litchi Flower Detection with Occlusion

Abstract

Controlling shoot and promoting flower of litchi is one of the key technologies for the high yield of fruit trees, but the occlusion caused by dense growth has a great impact on the detection of flowers. To solve the problem, this paper designs a polyphyletic loss function to detect the occlusive litchi flowers. A new aggregation loss term is proposed in the loss function to force the proposal box to approach and compactly locate the corresponding target. At the same time, the segmentation loss of the bounding box specially designed for the dense crop scene is added to keep the prediction box away from the surrounding objects and improve the robustness of detecting a large number of flowers. We conducted experiments on the self-built litchi flower data set and the multi species fruit flower data set at the same time. Compared with other methods, the proposed method has higher robustness and detection accuracy, which provides an important idea and method for flower number statistics and fruit yield prediction in dense scenes.