頁籤選單縮合
題 名 | 胴裂米反光特性的分析與應用=Image Analysis and Application of Reflection of Cracked Rice |
---|---|
作 者 | 萬一怒; 林建銘; 廖家興; 周瑞仁; | 書刊名 | 農業機械學刊 |
卷 期 | 6:3 1997.09[民86.09] |
頁 次 | 頁89-99 |
分類號 | 434.119 |
關鍵詞 | 胴裂米; 機械視覺; 影像處理; 品質檢測; Cracked rice; Fissured rice; Image process; Machine vision; Quality inspection; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究探討以機械視覺檢測胴裂米所需的最佳打光條件,並發展一簡單有效的胴 裂米檢測演算法。胴裂米以側向打光,可使胴裂面因光的反射而在其左右兩側有明顯的亮度 差異。此差異愈明顯則胴裂米檢測的雜訊就愈小,並使得檢測的速度與精度增加。實驗顯示, 上方單側打光時,打光角度5度左右胴裂線最明顯,當打光角度逐漸增加到30度,其胴裂 線明顯度隨著快速遞減,而從30度到55度打光角度之間胴裂線明顯度變化較小,55度之 後又快速的降低。下方打光時,胴裂線在打光角度20度至40度之間最明顯。台農67號糙 米粒,表面藍色光灰階值比紅、綠色光灰階值低,但是三者灰階梯度變化卻十分相近,因此 胴裂線的偵測採用單色光即可。米粒胚芽朝向或背向光源雖會造成米粒影像邊緣灰階變化, 但是胴裂多出現在胚乳部位,故對判別胴裂米的影響不大。本研究所發展的胴裂米搜尋檢測 演算法,應用搜尋框計算米粒長軸方向左右灰階的差值,使用486-33MHz個人電腦,只需 約0.087秒即可完成一粒米的搜尋檢測。經使用台中農業改良場所提供人工選別出的胴裂糙 米樣本測試,其結果正確率可達100%。 |
英文摘要 | Machine vision is applied in this research to study an optima lighting condition of detecting cracked rice kernels. Based on the study, a simple and efficient crack-line searching algorthm is proposed. If a cracked kernel is lighting from one side, the reflection of the crack-line makes two different brightness sections on left and right sides along the line. The larger the difference the less image noise need to be dealt with while processing the quality inspection. A larger gray level difference also increases inspection speed and precision. According to the test, lighting a kernel from one side with angle above horizontal and around 5。 ,the clearest crack-line can be found. The clearness reduces linearly while the lighting angle increasing from 5。to 30。. The clearness of the crackline reduces rapidly after lighting angle over 55。. When kernel-lighting is from one side under horizontal, the largest gray level difference of a crack-line can be found within the angle range of 20。 to 40。. According to the surface color image analysis of "Tainung 67" brown rice, the blue gray level is less than the green and red gray levels. However, the gradients of red. Green and blue gray level are similar, which show that monochromatic image is adequate to detect kernel crack. In image analysis, the embryo of a kernel toward or backward the lighting source will cause different gray level change on boundary. However, the crack-line usually occurs on inner section of endosperm. Therefore, the direction of embryo does not have an effect on the inspection. An algorithm proposed in this research takes about 0.087 sec to complete a kernel' s inspection by using a 486-33 MHz personal computer. It applies a rectangular searching frame scanning along kernel's long axis to find a crack-line. In the test, the correct rate of inspecting cracked kernels samples prepared by "Taichung District Agricultural Improvement Station" can reach 100%. |
本系統中英文摘要資訊取自各篇刊載內容。