| 摘要: | 生物辨識特性可減少遺失辨識物品，例如身分證或員工證…等，還有忘記帳號或密碼。近年來生物特徵辨識不斷有開始研究，並有許多廣泛應用於生活上，例如：臉部辨識、指紋解鎖、虹膜辨識、掌紋識別…等，本篇重點在於掌紋相關研究。相較於傳統式掌紋辨識方法大多是接觸式辨識或擷取掌紋，會造成使用者不方便和實用性。為了改進這個問題，本研究提出非接觸掌紋偵測系統，首先用智慧型手機分別在不同時間與地點拍攝影像，再使用labelImg標註掌紋感興趣區域，接著YOLOV3和YOLOX卷積神經網路（CNN,Convolutional Neural Network）進行模型訓練，訓練完成後以mAP模型評估指標，YOLOX為1.0、YOLOV3為0.995。本研究設計出掌紋偵測，在少量學習樣本，也具備良好模型評估結果，並能夠在不同的環境下進行掌紋區域偵測。;Biometrics reduce lost identification items, such as identity cards or employee cards... and forget your account number or password. In recent years, biometrics have been continuously studied, and many are widely used in life, such as: face recognition, fingerprint unlocking, iris identification, palm recognition... And so on, this article focuses on palm print-related research. Compared with the traditional palm print identification methods are mostly contact-type identification or palm print, will cause users inconvenience and practicality. In order to improve this problem, this study proposed a non-contact palm detection system, first with a smartphone at different times and places to take images, and then use label AppleImg to mark the area of interest in palm print, and then YOLOV3 and YOLOX convolutional neural network model training, after training with the mAP model evaluation index, YOLOX 1.0, YOLOV3 for 0.995. This study designed palm grain detection, in a small number of learning samples, but also has a good model evaluation results, and can be detected in different environments palmprint area detection. |
