水果影像辨識分級系統之建置—以番石榴為例

The Construction of Image Identification system for fruit grading processes — An Example for Guava

吳縈蓁(Wu Ying-Jhen)

指導教授 ： 黃怡詔

國立屏東科技大學/管理學院/工業管理系所/碩士(2020年)

https://doi.org/10.6346/NPUST202000030

摘要

台灣的水果豐富多樣，除了自給自足外，外銷市場也是不容小覷，因此水果的分級是相當重要的，現今大多使用人工目測將水果依照外觀品質做分級，除了處理成本高之外，每個人主觀判別的標準也不依，因此本研究利用MATLAB撰寫處理程式，開發以影像辨識番石榴外觀的非破壞性品質檢測應用程式，目的在於將水果快速且有系統地分級，根據蒐集樣本設定特徵值，並找出判斷番石榴外部品質之標準。同時也探討拍攝番石榴樣本的實驗環境，實驗結果顯示在白色背景的自然光下所拍攝的初始影像，去除背景及影像辨識的效果相較於其他環境較佳，運用影像處理技術找出番石榴的特徵值，最後將判斷結果彙整成Excel紀錄表單，讓使用者能快速檢視缺點區域百分比，其經系統實際測量後，此辨識系統平均每判斷一張影像需花費4秒時間。除了提供農民一種快速檢測水果分級缺點的參考標準，而缺點的特徵值設定，可彈性地依據使用者的判別標準提高或降低特徵值，以達辨識分級的目的。提供消費者值得信賴的好品質，建立與塑造品牌蔬果形象，創造優質精緻化農業。

關鍵字

影像辨識 ； 水果分級 ； 非破壞性品質檢測 ； 快速檢測

並列摘要

Taiwan's fruits are rich and diverse. In addition to being self-sufficient, the export market is also not to be underestimated. Therefore, the classification of fruits is very important. Today, most people use manual visual inspection to classify fruits according to their appearance quality. In addition to high processing costs, everyone is subjective. Discrimination standards are not adhered to, so this research uses MATLAB to write a processing program to develop a non-destructive quality detection application that uses the image to identify the appearance of guava. The purpose is to quickly and systematically classify fruits, set characteristic values based on collected samples, and Find out the criteria for judging the external quality of guava. At the same time, the experimental environment for taking guava samples is also discussed. The experimental results show that the initial image was taken under natural light with a white background, the effect of removing the background and image recognition is better than other environments. Use image processing technology to find out the guava's Feature values, and finally, the judgment results are aggregated into an Excel record form, allowing users to quickly view the percentage of defective areas. After the system actually measures, the recognition system takes an average of 4 seconds to judge each image. In addition to providing a reference standard for farmers to quickly detect the shortcomings of fruit grading, the characteristic value setting of the shortcomings can flexibly increase or decrease the characteristic value according to the user's discrimination criteria to achieve the purpose of identifying and grading. Provide consumers with trusted good quality, establish and shape the brand's image of fruits and vegetables, and create high-quality and refined agriculture.

並列關鍵字

image recognition ； fruit grading ； non-destructive quality testing ； rapid testing

參考文獻

參考文獻

中文部分

1.王至誠(民107)。基於圖像識別技術的電子元器件分類。電腦編程技巧與維護，1，142-147。

2.張佑瑋(民94)。自動灰階影像上色系統。台灣科技大學資訊管理系，碩士論文。

3.王維、宋雪樺(民106)。基於動態閾值的彩色圖像邊緣檢測與分析。信息技術，8，204-207。