類神經網路應用於脈衝輕擊番石榴之分級研究

Grading Maturity of Guava for Pulse Slight Impact Using a Neural Network

連振昌(Cheng-Chang Lien) ； 林宏駿(Hong-Jyun Lin) ； 楊朝旺(Chao-Wang Young)

《台灣農學會報》 12卷1期 (2011/02) Pp. 1-17

https://doi.org/10.6730/JAAT.201102_12(1).0001

摘要

本研究是利用輕微碰擊試驗的方式，對水果成熟度進行非破壞性分級檢測，自行設計組裝一套小鋼珠自由掉落碰擊裝置對番石榴進行輕微碰擊，由力量感測器得到的力量時間波形圖可定義出各種碰擊相關參數，以番石榴的彈性常數作為其成熟度指標，由逐步迴歸分析選擇有顯著性差異的碰擊相關參數，最後應用類神經網路來探討番石榴成熟度等級的分級準確率。由試驗分析結果，使用四個顯著性差異的分級碰擊相關參數作為輸入，過成熟、半成熟及未成熟三種成熟度等級作為輸出，進行類神經網路分級檢測，當訓練1500次後，其訓練組的分級準確率為85.3%，驗證組的分級準確率為75.9%；訓練2000次後其訓練組的分級準確率為88.2%，驗證組的分級準確率為77.8%。由穿刺試驗結果可驗證輕微碰擊分級番石榴成熟度是一種非破壞性檢測。

關鍵字

輕微碰擊 ； 番石榴 ； 成熟度 ； 類神經網路

並列摘要

The slight impact test was used to grade the maturity of guava in non-destructive detection for this research. A steel ball drop free falling device was designed to estimated the maturity of guava using the slight impact test. The force-time waveform under impact was acquired through the load cell senor. These impact parameters were obtained after defining and analyzing the force-time waveform. The elastic constant of guava was used as a classification index. First, four significantly different parameters were found out by step regression analysis as the input layer the artificial neural network. Three maturity levels of over-ripe, half-ripe and un-ripe associated with the elastic constant of were guava was used as the output layer. Then then the accuracy of maturity classification was measured by the artificial neural network method. The results show that the accuracy of grading guava is 85.3% and 75.9% for the corresponding to training set and valid set after artificial neural network training of 1500 times. The grading accuracy is 88.2% and 77.8% for the corresponding to training set and valid set after training of 2000 times. The puncture test result proved the slight impact test for grading guava is a non-destructive method of detection.

並列關鍵字

Pulse slight impact ； Guava ； Maturity ； Neural network