Realization of Chrysanthemum Harvesting Recognition System based on CNN

Abstract

To realize the real-time identification, classification, and positioning of chrysanthemums under complex conditions such as light, occlusion, and posture by picking robots. This paper adopts an image recognition classification and positioning method based on CNN. First, an image training data set is established, and adaptive anchor frame calculation is used to locate the chrysanthemum data augmentation quickly. Finally, extract the features of the image, complete the model's training, and output the best feature model file. Finally, statistical data analysis is carried out for the multi-target chrysanthemum identification experiment. The experimental results show that the training model's recognition accuracy and recall rate in complex environments can reach 98% and 95%, respectively, and the model size is 14. 5MB. Compared with other algorithms, the accuracy and model size have certain advantages. The practicability of the model is verified, and the research can realize the rapid localization and classification of chrysanthemums by a picking robot.