Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/2684
標題: 以影像處理與類神經網路模式來進行柴油車黑煙判讀
Gauging the Black Smoke of a Diesel Car with the Image Process and Neural Network
作者: 李伯滄
關鍵字: 柴油車黑煙排放;影像處理;碎形維度;類神經網路
出版社: 機械工程學系
摘要: 
黑煙是柴油引擎汙染排放中最明顯的一部份,由於可用肉眼明顯的分辨出來,為一般民眾所厭惡。目前我國柴油車的排煙檢測方法採用目測及儀器檢測兩種方法,但此兩種方法都有其缺點存在。
為了更真實檢測柴油車的排煙,本研究提出〝以數位攝影來記錄柴油車的黑煙排放情況,並以影像處理技術來判斷黑煙的濃度〞的構想,因為數位攝影可以即時反應車輛真實排放情況,影像處理則可以消除主觀判斷因素。以下介紹本文主要研究方向。
黑煙影像辨識:本研究分別提出以動態法和靜態法來進行黑煙影像辨識,結果顯示兩法皆能完整的將黑煙影像由攝影所得的圖像中獨立出來。
黑煙影像特徵:本研究找出五組能夠充分反應黑煙排放影像的特徵,分別為黑煙碎形維度,黑煙平均灰階值,黑煙灰階值標準差,黑煙面積百分比和黑煙Euler-Number。
黑煙影像類神經網路模式:訓練出兩組類神經網路分別負責黑煙影像的判斷和黑煙煙度的預測。最後網路訓練結果為判斷黑煙影像誤判率達到0.667%;黑煙煙度預測結果R-square值達到0.614,誤判率為22%。

The emission of black smoke from diesel cars have always been complained by people for it's visible. Currently, there are two ways to detect the emission, one is to employ trained professionals for their judgement and the other is to utilize monitoring equipments for sampling analysis.
For speed a better results, we propose a new approach which is recording the discharge by CCD camera and determining consistently the opacity by image process. With this new approach, we can not only response the emission situation on real time but also ascertain the pollution problem more objectively. The major contributions of the research are:
1. Identification of black smoke's image: We use both continual and interval approaches to carry the work out, and the results seem to be satisfied.
2. Characterization of black smoke's image: Among many features we totally find five characteristic to characterize an the image of black smoke, these are fractal dimension, average value of gray level, standard deviation, area percentage and Euler number.
3. The neural network of black smoke's image: Two neural networks have been trained to find if the images belong to black smoke or not and gauge the components of white smoke respectively. Case studies show that the incorrect judgement of both neural networks is 0.667% and 22% in sequcent.
URI: http://hdl.handle.net/11455/2684
Appears in Collections:機械工程學系所

Show full item record
 

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.