Tea bud detection based on multi‐scale convolutional block attention module

Abstract For the sake of effectively identifying tea buds and improving the picking precision of mechanical picking, as a matter of fact, the traditional object detection algorithm has some problems such as poor detection effect and robustness under unstructured environment, the application of Yolo‐Tea object detection algorithm in tea bud detection under unstructured environment was explored. First, by introducing multi‐scale convolutional block attention module (MCBAM) and multi‐scale prediction layer into a network, the model gathers important information that is beneficial to tea buds classification and enhances the detection of small object tea buds in the dense scene. Then, based on the specific tea buds dataset, the anchor boxes are re‐clustered using K‐means and genetic algorithm. Finally, EIoU loss function is introduced into the boundary box regression stage to reduce the missed detection and speed up the convergence of the model. The multi‐detection box generated by the object is suppressed by soft‐non‐maximum suppression, the test effect image has a final boundary box, and the category score is output. The experiment results show that the mAP of the proposed algorithm for tea buds is 95.2%. Compared with the common object detection algorithm, the network shows superior performance in tea buds detection, which can effectively improve the recognition effect of tea buds under an unstructured environment.