- 學位論文
建立超分光光譜影像分辨國產合格與不合格雞隻屠體檢測技術之研究
Development of Detecting Technique for Wholesome and Unwholesome Domestic Chicken Carcass by Ultra-spectral Image
摘要
光譜影像(Spectral image)檢測為一種非破壞性及快速檢測的方法，它可同時提供光譜及影像的資訊，近年來被廣泛使用在農產品內部品質及外部品質檢測上。為開發國產肉雞的自動檢測系統，本研究利用本實驗室自行開發的超分光光譜影像系統(解析度達0.3 nm，波長範圍400~1100 nm)，以線掃描方式拍攝國產肉雞屠體胸部光譜影像，波長範圍404~750 nm；本研究的肉雞樣本30隻為市場上所購置的16隻合格樣本及肉雞屠宰場經獸醫師檢查不合格的樣本14隻，分別拍攝這些肉雞樣本的線光譜影像，然後組合成單一波長光譜影像，並擷取胸部區域影像光譜強度進行分析，結果使用620 nm與573 nm兩波長之強度比值，得到不合格與合格肉雞之正確率分別為100%及 87.5%；另外，使用573 nm波長影像，將合格與不合格雞隻的影像紋理進行13種特徵擷取，並將13種紋理特徵及620 nm、573 nm兩波長之強度比值使用單層感知器進行訓練及測試，測試結果合格與不合格肉雞之判別率分別為100％及85.8％。本研究方法可為自動檢測系統之研發提供良好參考。
並列摘要
Spectral image is a non-destructive and rapid detection method, which provides both spectral and image information. In recent years, spectral image has been widely used in internal and external quality evaluation for agricultural products. This study aims to develop an automatic detection technique for chicken carcasses by using a home-made ultra-spectral imaging system (resolution of 0.3 nm at range 400 ~ 1100 nm). Sample of chicken carcass were scanned at chest to obtain its line image at wavelength 404 to 750 nm. Totally 30 chicken samples were tested that include 16 wholesome and 14 unwholesome samples. Wholesome chickens were purchased from market while unwholesome samples were taken from slaughter house and examined by veterinary. Line images of chicken samples were registered into spectral images and were analyzed based on spectral intensity of chest area. Results showed that by using intensity ratio at 620 nm and 573 nm, the detection accuracy for wholesome and unwholesome were 100% and 87.5%, respectively. In addition, spectral images at 573 nm were used to extract 13 texture features, along with band ratio feature, formed input features for a single-layer perceptron network. Test result showed that accuracy was 100% and 85.8% for wholesome and unwholesome samples, respectively. The proposed method provided a good reference for development of automatic detection system.