| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 陳仕軒 | zh_TW |
| DC.creator | Shih-Hsion Chen | en_US |
| dc.date.accessioned | 2014-8-22T07:39:07Z | |
| dc.date.available | 2014-8-22T07:39:07Z | |
| dc.date.issued | 2014 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=101522040 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 因為環保議題愈來愈受重視的緣故，近年來有愈來愈多有關太陽能發電的研究。由於太陽能發電的效能主要仰賴地表所接收到的輻射量多寡，而天空中的雲朵是影響地表接收輻射最大的因素，因此若能對天空中的雲朵做即時性的觀測，分析目前主要分佈的雲層種類，便能藉此預測短期日射量的變化，以幫助太陽能電廠做電力上的調配與控管。但雲朵分類是項耗時且需長期觀測累積經驗方能確保正確性的工作，因此近來有人利用影像處理的技術對天空的雲朵進行分析，以幫助電廠自動化地瞭解未來某段時間內太陽受雲朵影響的程度，進而判斷是否需要採取其他相應措施。
然而，正如所有分類的問題會遭遇到的困難，雲朵的分類也存在著相當高的難度：除了相同種類的雲外型千變萬化，不易辨識，不同種類的雲有時也可能非常相似，而導致誤判。此外影像也容易因為受到太陽的高度曝光而使影像產生大面積且模糊的亮點，而使原本無雲的部分被電腦誤判為含有大片的雲朵，或是突然被遮蔽而使整個畫面突然偏暗，這些都是可能影響分類的重要因素。
為解決上述問題，本篇論文我們提出以區塊代替全影像進行分類的方法。首先，我們將全影像區分成數個等大的區塊，並從中取出統計特徵（Statistical Features)與區域紋理特徵（Local Texture Features)，再利用最近鄰居（k-Nearest Neighbor)與支持向量機（Support Vector Machine)兩種方法分別對此兩類樣本個別訓練出分類模型。之後對影像中的區塊逐一分類，並且利用投票的方式決定該影像的最後分類結果。最後本實驗將顯示利用分割區塊法可以有效地降低多種雲同時出現的情形所造成的誤判，提高分類準確率。 | zh_TW |
| dc.description.abstract | With the increasing importance of environment protection, there are more and more research works in analyzing the clouds to help the solar plant to realize the effect caused by the clouds after a period of time. Because the ability to block the Sun differs from different kinds of clouds, automatically classifying the clouds will help the solar plant to allocate or manage the power system.
However, there are some difficulties existing in cloud classification. The large variety within the same cloud type and the similarity between the different cloud types both make the task more challenging. Besides, the dramatic light change caused by the relative position of the Sun and clouds will make it more difficult.
To deal with the problems, we make efforts in extracting more powerful features to classify different cloud types. Besides, we propose a method which use ＂block＂ instead of whole image to classify the mixed conditions. In our research, we divide the whole image into multiple equal-sized blocks, then extract statistical and local texture features from them. Then we’ll train models using k-Nearest Neighbors and Support Vector Machine. Before we start classify, we use Solar Position Algorithm to skip the blocks where the Sun locates in. After classification of each block, we use voting mechanism to decide the result of the image. In our experiment, we found that the proposed method outperforms the method using only whole image. | en_US |
| DC.subject | 雲分類 | zh_TW |
| DC.subject | 太陽位置偵測 | zh_TW |
| DC.title | 在全天空影像中使用紋理特徵之雲分類 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Cloud Classification Using Texture Features in All-Sky Images | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |