文献类型：期刊文献

英文题名：A-Pruning: A Lightweight Pineapple Flowers Counting Network Based on Filter Pruning

作者：Yu, Guoyan[1,3]; Cai, Ruilin[1,2]; Luo, Yingtong[1,2]; Hou, Mingxin[1,3]; Deng, Ruoling[1,3]

机构：[1] School of Mechanical Engineering, Guangdong Ocean University, Guangdong Province, P.R, Zhanjiang, 524088, China; [2] Guangdong Marine Equipment and Manufacturing Engineering Technology Research Centre, Guangdong Ocean University, Guangdong Province, P.R, Zhanjiang, 524088, China; [3] Southern Marine Science and Engineering Guangdong Laboratory, Guangdong Province, P.R, Zhanjiang, 524088, China

年份：2022

外文期刊名：SSRN

收录：EI(收录号：20220332439)

语种：英文

外文关键词：Antennas - Deep learning - Learning systems - Signal detection

外文摘要：Abstract: During pineapple cultivation, detecting and counting the number of pineapple flowers in real-time and estimating the yield is essential. Compared with traditional manual detection, the deep learning methods are more efficient in real-time performance. However, existing deep learning models are characterized by low detection speeds, which cannot be applied in real-time on mobile devices. This paper presents a lightweight model in which filter pruning is used to compress the YOLOV5 network. It introduces an adaptive Batch Normalization layer evaluation mechanism during the pruning process to evaluate the performance of the subnetwork. By this approach, the network with the best performance can be found quickly after pruning. Then, an Efficient Channel Attention mechanism is added for the pruned network to constitute the new YOLOV5_E network. The results show that the YOLOV5_E network proposed in this paper only loses 0.9% accuracy, the number of parameters remains 24.2% of the original, the model size is 3.8MB, and it can achieve a running speed of 178 frames per second. YOLOV5_E network has higher accuracy and detection speed by comparison with the other five networks, showing favorable outcomes in both accuracy and real-time performance. The combination of YOLOV5_E and StrongSORT enables mobile devices' real-time detection and counting. It has been validated on the NVIDIA Jetson Xavier NX development board with average detection speeds up to 24 frames per second. The proposed YOLOV5 _E network can be effectively used in agricultural equipment such as Unmanned Aerial Vehicles, providing technical support for detection and counting crops on mobile devices. ? 2022, The Authors. All rights reserved.