查詢結果分析
相關文獻
- 類神經網路應用於影像處理魷魚分級之研究
- Neural Network Procedures for Taguchi's Dynamic Problems
- 應用類神經網路預測東南亞金融風暴期間臺幣對美元的短期匯率
- 製程間具相關數據之趨勢模型偵測探討
- 手寫數字辨識模式之建立--結合遺傳演算法與類神經網路
- Form Segmentation and Component Classification for Clinic Document Image Analysis
- 類神經網路應用於臺灣股市預測:統合基本面與技術面資訊
- 倒傳遞類神經網路在波浪時序列預報之應用
- 類神經網路在集水區降雨逕流模擬之應用
- 達成中央空調舒適度與省能控制之研究
頁籤選單縮合
題 名 | 類神經網路應用於影像處理魷魚分級之研究=Study of Squid Fishing Image Grading by Neural Network |
---|---|
作 者 | 郭興家; 吳立仁; 張允豪; | 書刊名 | 農業機械學刊 |
卷 期 | 8:1 1999.03[民88.03] |
頁 次 | 頁25-38 |
分類號 | 483.5 |
關鍵詞 | 魷魚分級; 類神經網路; 倒傳遞; Grading of squids; Neural network; Back-propagation; |
語 文 | 中文(Chinese) |
中文摘要 | 遠洋魷釣作業為我國漁業發展重點,傳統魷釣船上的魷魚分級皆利用人工分類, 不僅所需人力甚多,且漁民長時間工作、效率自然減低。為了解決此一困境,本研究提出利 用類神經倒傳遞網路理論、應用影像處理技術作魷魚分級,以減少人力,並促進魷釣業的自 動化。首先利用影像處理擷取魷魚原始影像,研究發現以40W四組燈泡為光源,結合門檻 值225所處理影像效果最佳,並利用遮罩將魷魚邊緣強化,以尋求邊緣,由八相都鍵結法可 得到魷魚周長,計算出形心,並進一步求得最長軸、寬度、長度。如此將魷魚的特徵擷取出, 共計有面積、周長、最長軸、長度、寬度等。並嘗試將十組特徵值,組合搭配成A、B、C 三組模型,當作輸入向量,輸入類神經網路,經神經元隱藏層,其輸出可將魷魚分為四等級, 依所建模型可得正確率92%之分級結果,此項結果可配合分級機周邊設備做影像分級之參 考。 |
英文摘要 | The deep-sea squid fishing is one of the most important parts of our fishery. Traditionally, the grading of squids on ship is done manually. It is not only labor-intensive, but also wasting time. As a result, it decreases the efficiency of grading. In order to improve the efficiency of squid grading on ship, we propose a new method, which combines the theory of artificial neural network and the technique of image processing to make the grading of squids automatically. The grading is divided into two parts, one is adopting the image processing of the original image of squid, and the other is completing the grading by Back-Propagation neural network. By utilizing the technique of image processing to get the characteristic items of the squid, including the area, the center of the area, the perimeter, the width, and the length of the longest axis. In our research, the optimal result can be obtained by the source of 4(40 watt)each light-bulbs and adopt the image processing which use the threshold number of 225. By combining these with ten characteristic values to build three models (A,B,C). The three models are the input vectors of the Back-Propagation neural network, and then we might obtain four different grading levels as the output results. The validity of the grading model can reach 92%. As the same time, the results could be the references and cooperate with other equipment to analyze the processes of squid grading. |
本系統中英文摘要資訊取自各篇刊載內容。