水果智慧型選別之研究

Intelligent Algorithm for Fruit Sorting

張文宏(Wen-Hung Chang) ； 陳世銘(Suming Chen) ； 連豐力(Feng-Li Lian) ； 許豐益(Feng-Yih Hsu) ； 謝廣文(Kuang-Wen Hsieh)

《農業機械學刊》 3卷4期 (1994/12) Pp. 25-35

https://doi.org/10.30062/JAM.199412.0003

摘要

本研究之主要目的在於利用模糊理論及倒傳遞類神經網路模擬水果之選別，並探討不同網路之收斂速度及選別正確率。選別模式採用倒傳遞類神經網路及模糊類神經網路二種。以後者而言，利用模糊集合理論，將影響水果等級的因素，包括水果投影面積大小、重量、平均灰度値等因子以歸屬函數的方式來描述，將人類專家的經驗法則整理為模糊規則；再利用類神經網路非線性組合及自我學習的特性，來調整歸屬函數的形狀，經三層（輸入層、隱藏層及輸出層）的倒傳遞類神經網路之訓練而得到各影響因子對選别指標不同的影響程度。實際以檸檬進行選別試驗，倒傳遞類神經網路的選別較接近人工選別，但收斂速度較慢；模糊類神經網路的收斂速度快，且對選別結果較具强健性，不過選別正確率較僅用倒傳遞類神經網路為差。

關鍵字

模糊理論 ； 類神經網路 ； 模糊類神經演算法 ； 水果選別

並列摘要

In this study, fruit sorting by using fuzzy theory and error-back-propagation (EBP)neural network was investigated, the convergence rate and sorting accuracy were discussed; the performance of EBP neural and fuzzy-neural networks were evaluated. Regarding fuzzy-neural sorting algorithm, the membership functions were developed for fruit's properties including projected area, weight, averaged gray level and etc.; and fuzzy rules were derived from human's experience; the membership functions were then adjusted by using self-learning scheme of neural network, and an EBP network was further imposed to obtain the optimum weights for the fruit's properties related to sorting criteria. Lemons were used to study the established networks. The algorithm of EBP network gave better sorting accuracy comparing with fuzzy-neural network, however, fuzzy-neural network showed its robust ability in sorting and had a much faster convergence rate.

並列關鍵字

Fuzzy ； Neural network ； Fuzzy-neural algorithm ； Fruit sorting