| 摘要： 本研究使用數位信號處理技術與統計鑑別分析，以發展農產品品質非破壞撞擊檢測方法及品質檢測指標之建立模式。設計ㄧ單擺撞擊裝置，撞擊與量測農產品之反應信號，藉由檢測信號之振幅頻譜、實部頻譜、虛部頻譜、撞擊力-時間曲線之斜率與曲率及曲線上之微小震盪信號，推導出與農產品品質變化有關的撞擊檢測參數。經變異數分析檢定（信賴區間95%），獲得振幅頻譜之有效頻率，及實部頻譜與虛部頻譜中振幅等於零之頻率與振幅最大值、最小值及其相對應之頻率等與品質變化之關係。同時原始撞擊信號經3階數位低通濾波器平滑化處理與有限差分計算，得到撞擊力曲線之最大斜率、最小斜率、最大曲率、最小曲率、第一轉折點時間與第二轉折點時間等之品質量測參數。進ㄧ步研究發現質地柔軟之農產品可應用Wiener-Khintchine定理，計算撞擊力曲線上之微小震盪信號的功率頻譜密度(PSD)，探討其品質之變化。許多參考研究使用單一撞擊檢測參數分類番石榴、芒果、牛心番茄的品質等級及雞蛋種類之辨識，其準確率不超過70%。然而結合時域與頻域等各類型撞擊檢測參數，經統計之鑑別分析做部分參數的最佳化組合，形成農產品品質檢測指標，可獲得較高準確度的檢測結果。研究顯示，對於番石榴成熟度、芒果成熟度、芒果酸度、牛心番茄成熟度的品質分類及雞蛋種類的辨識，分別可達到82.7%、81.0%、85.7%、80.0%、82.5%的準確率。This study develops a nondestructive inspection method and quality index construction model to evaluate the quality of agricultural products using digital signal processing and statistical discriminate analysis. A pendulum device is designed to impact and measure the response signal of products. The impact parameters which correspond to the change in the quality of agricultural products are obtained from the amplitude spectrum, real-part spectrum, imaginary-part spectrum and the slope, curvature and micro-fluctuation signal of the impact force-time curve. The analysis of variance (95% confidence interval) is used to determine the effective frequency and amplitude of the spectra of amplitude, real-part and imaginary-part as the inspection parameters for the quality of agricultural products. Analyses indicate that a three-order lowpass digital filter can smoothen the raw impact force-time curves to calculate their exact slope and curvature using finite difference. The maximum and minimum slopes, maximum and minimum curvatures and time of inflection point are valid impact parameters from the curve. Additionally, the power spectral density of the micro-fluctuation signal obviously reflects the variation in the texture of soft products using the Wiener-Khintchine theorem. The accuracies are lower than 70% using an impact parameter to classify the quality of guavas, mangos and tomatoes, as well as to egg variety. However, the classification accuracies can be improved by more than 10% when using high accuracy indices with parameters selected from the time and frequency domains, as well as their combinations, which are established using statistical discriminate analysis. Test results demonstrate that the accuracy reaches 82.7%, 81.0%, 85.7%, 80.0% and 82.5% in quality classification of guava maturity, mango maturity, mango acidity and tomato maturity and recognition classification of egg variety, respectively. |
頁面載入中請稍後...