洋香瓜糖度檢測之研究－（一）影像紋理分析法

Determination of Sugar Content in Cantaloupe-(Ⅰ) Image Texture Analysis

郭立穎(Li-Ying Guo) ； 陳世銘(Su-Ming Chen) ； 張文宏(Wen-Hung Chang)

《農業機械學刊》 7卷1期 (1998/03) Pp. 75-86

https://doi.org/10.30062/JAM.199803.0006

摘要

In this study,an image processing based surface texture analysis was usedto predict the sugar content and released aroma of cantaloupe.The digitalimage of cantaloupe surface was grabbed to form a co-occurrence matrix,andthen seven texture parameters,ASM,CON,COR,LH,E,CS and CP,werecalculated.The analysis of the relationship between sugar content and textureparameters indicates that the larger values of ASM and LH,or the smallervalues of CON and E are,the higher sugar content of cantaloupe is.Amongthe linear regressions with texture parameters,the highest values of correlationcoefficient are -0.887 for sugar content and 0.478 for aroma.The highest correlation coefficient is -0.902 for sugar content in non-linear regression analysis.The LH texture parameter is the most correlated with sugar content among alltexture parameters for both linear and non-linear regressions.Neural networkanalysis was also used to predict the sugar content,and a correlation coefficientof 0.911 was obtained.

本研究設計製造一組水果選別裝置，由推桿機構與影像處理系統所組成。利用推桿機構使水果滾動，並運用CCD攝影機連續取像，每一次可同時對兩排-每排各有3個-共6個之水果擷取影像，連續取三張像，經影像處理後即可計算最前端兩個水果之平均周長值，依序連續取像，並持續計算水果之平均周長值。本研究選擇蘋果與柳丁做為試驗對象，根據試驗結果顯示，推桿機構以2.1 cm／sec速率前進，每粒水果完成周長計算的平均時間為3 sec，而周長所計算之平均精確度：蘋果為95.9％，而柳丁可達97.0％。

A fruit sorting machine that consists of a push-rod mechanism and animage processing system had been developed in this study.The image of fruitwas grabbed by a CCD camera while the fruit was rolling due to the push-rod mechanism.The CCD camera could grab two rows-each row contained3 fruits-that was a total of 6 fruits in one image.The perimeters of twofruits in the front of two rows could be evaluated after completing the imageprocessing of three consecutive images.Two kinds of fruits,apple and orange,were used to test the machine.According to the results,when the speed ofthe push-rod mechanism was 2.1 cm/sec,the average time spent on calculatingfruit perimeter is about 3 sec,and the precision in average is 95.9% for apple,and 97.0% for orange.

本研究利用近紅外線光譜檢測洋香瓜果汁及果肉之糖度，探討之波長範圍為1000 nm至2500 nm，而糖度範圍為5.5至13.2°Brix。在果汁透射光譜分析中，2274 nm，2241 nm，2321 nm三個波長不論是在原始光譜(log(1／T))或二次微分光譜(D^(2)log(1／T))中，均對洋香瓜之糖度預測有極大的影響；由二次微分光譜2274 nm，1143 nm，2241 nm以及2321 nm四個波長組合之校正方程式對預測樣本之相關係數可達0.970，而標準預測誤差(SEP)及相對預測誤差(RSEP)為0.383及3.64％。果肉反射光譜之原始光譜對於糖度的預測結果並不理想，但是其二次微分光譜(D^(2)log(1／R))與其糖度值則有極佳的線性關係；試驗結果顯示，由1688 nm，2321 nm，1401 nm，2000 nm及1106 nm五個波長所組成的多重線性迴歸校正方程式，其標準校正誤差(SEC)為0.728，相關係數為0.917，對於預測樣本的預測，其標準預測誤差(SEP)及相對預測誤差(RSEP)分別為0.769及7.21％，相關係數則為0.892。不論果汁或果肉之糖度檢測，二次微分光譜顯然較原始光譜為佳，尤其以單波長校正方程式更為明顯。

關鍵字

紋理分析 ； 糖度 ； 洋香瓜 ； 影像處理 ； 共生矩陣 ； 香味

並列摘要

In this study, an image processing based surface texture analysis was used to predict the sugar content and released aroma of cantaloupe. The digital image of cantaloupe surface was grabbed to form a co-occurrence matrix, and then seven texture parameters, ASM, CON, COR, LH, E, CS and CP, were calculated. The analysis of the relationship between sugar content and texture parameters indicates that the larger values of ASM and LH, or the smaller values of CON and E are, the higher sugar content of cantaloupe is. Among the linear regressions with texture parameters, the highest values of correlation coefficient are -0.887 for sugar content and 0.478 for aroma. The highest correlation coefficient is -0.902 for sugar content in non-linear regression analysis. The LH texture parameter is the most correlated with sugar content among all texture parameters for both linear and non-linear regressions. Neural network analysis was also used to predict the sugar content, and a correlation coefficient of 0.911 was obtained.

並列關鍵字

Texture analysis ； Sugar content ； Cantaloupe ； Image processing ； Co-occurrence matrix ； AromaTexture analysis ； Sugar content ； Cantaloupe ； Image processing ； Co-occurrence matrix ； Aroma