基于GA?BP神經網絡的手勢識別精度優化研究

郭興 徐武 唐文權

摘? 要： 針對5DT數據手套手勢識別過程中存在的精度問題，傳統的BP神經網絡算法受到其自身因素的影響，導致出現輸出手勢缺失、變形、精度差的問題。為此，該文提出一種GA?BP權值優化算法，能有效克服BP算法局部尋優的缺點，使輸出值不斷地接近期望數值，防止陷入局部極小的情況，可以克服輸出圖像缺失、變形的問題。在GA?BP算法的基礎上，對函數輸出誤差的最大值進行權值優化，解決輸出手勢精度差的問題。實驗結果表明，基于GA?BP神經網絡權值優化算法改善了手勢識別的精度。

關鍵詞： 手勢識別; 精度優化; GA?BP神經網絡; 權值優化; 效果分析; 算法仿真驗證

中圖分類號： TN915.06?34; TP391.9? ? ? ? ? ? ?文獻標識碼： A? ? ? ? ? ? ? ? ? ? ? 文章編號： 1004?373X（2020）06?0183?04

Research on gesture recognition accuracy optimization based on GA?BP neural network

GUO Xing， XU Wu， TANG Wenquan

（College of Electrical and Information Engineering， Yunnan Minzu University， Kunming 650500， China）

Abstract： For the accuracy problem in the process of gesture recognition of 5DT data gloves， the traditional BP neural network algorithm is affected by its own factors， resulting in the absence， deformation and poor precision of the output gesture. A GA?BP weight optimization algorithm is proposed， which can effectively overcome the shortcomings of local optimization of BP algorithm， make the output value close to the expected value continuously， prevent getting into the local minimum， and overcome the missing and distortion of output image. On the basis of the GA?BP algorithm， the weight optimization for the maximum value of the function output error is conducted to improve the precision of the output gesture. The experimental results show that the weight optimization algorithm based on GA?BP neural network can improve the accuracy of gesture recognition.

Keywords： gesture recognition; precision optimization; GA?BP neural network; weight optimization; image output; effect analysis; algorithm simulation verification