基於Android與雲端平台上完成蝴蝶辨識系統

The Implementation of an Android and Cloud Computing-based Butterfly Recognition System

黃家洋(Chia-Yang Huang)

指導教授 ： 蘇義明

義守大學/電機資訊學院/電子工程學系/碩士(2014年)

https://doi.org/10.6343/ISU.2014.00252

摘要

本篇論文探討如何利用Android手機之照相功能與網路連接的即時和便利性，並透過雲端運算，完成蝴蝶辨識系統。
首先在Android手機上執行客戶端的程式，利用此程式拍攝蝴蝶之畫面，之後可選擇是否上傳至雲端伺服器。上傳至伺服器後，伺服器會立即進行蝴蝶辨識系統執行辨識，此辨識系統以視覺注意力理論為基礎，此辨識系統中，前處理階段會先將圖像尺寸標準化，再將色彩空間轉換至LMS色彩空間，並將亮度標準化，減少光的亮度所造成的影響，接著使用訓練好的獨立成分分析濾波器(ICA Filter)找出圖像ICA特徵，並將此特徵放入視覺注意力模型(Visual Attention Model)產生顯著圖(Saliency Map)，最後透過簡單貝氏分類器與K個最近鄰居法，與訓練好的資料庫比較相似度，將最佳蝴蝶類別結果當作索引，在關聯式資料庫管理系統(MySQL)內，搜尋蝴蝶品種資料，最後將五種蝴蝶的品種、活動範圍、特徵描述等資料回傳至使用者的手機端。
實驗結果顯示，此辨識系統僅使ICA特徵，真實手機辨識率可達83%，此系統提供快速且穩定的效能，並由於將複雜的運算都交給雲端的伺服器，降低手機的規格配置需求並提升續航力，可以讓更多人使用本系統。

關鍵字

資料庫 ； 視覺注意力 ； 顯著圖 ； 辨識系統

並列摘要

This thesis discusses how to use the cloud-computing to accomplish a butterfly recognition system via the network connection and camera of Android phone.
First, the system is used to execute the client program on Android phone to take a butterfly image, then is chosen whether to upload to cloud server or not. After the image is uploaded to server with network connection, the server will execute the recognition processing using the theory of human visualization to achieve recognition processing. At pre-processing step, the image is normalized to specific image size and transformed from the RGB color space to LMS color space, and normalize the image luminance to reduce the influence of the light brightness. After that, we used Independent Components Analysis (ICA) filters to find the ICA feature and combine the visual attention model, to compute the saliency maps. In addition, the approaches of Naïve Bayesian classifier and K-Nearest Neighbor algorithm are used to accomplish the training and classification processing. After finding the top five butterfly’s candidates, the recognition system will return butterfly’s data from MySQL database to Android phone, such as butterfly’s name, range, distribution, description, etc.
The experimental results show that the recognition system uses only one feature, up to 83% recognition rate on Android phone testing. Furthermore, it can reduce the phone requirement and promote phone's endurance by using the cloud server to achieve the complex calculation. Finally, it will be helpful for more people to use this system to real-time inquiry the butterfly data in real world.

並列關鍵字

Android ； MySQL ； Visual Attention ； Saliency Map ； Recognition System

參考文獻

Phones for the First Time in 2013.

pp. 3741-3744, 2010

[37] Color vision, wiki pedia

[40] M. Levine, “Fundamentals of Sensation and Perception.”

[52] Leeds Butterfly Dataset