Evaluation of banana and pear fruit maturity stages using laser backscattering images, artificial neural network and support vector machine techniques

Consumers considered ripeness of fruit as a very important factor in making choices of purchase. Ripeness in fruit generally affects their eating quality and market price. Quality attributes of fruit determined the extent of its acceptability and satisfaction by the consumers. Many quality attributes have been used for the determination of fruit quality, among them are colour, firmness and soluble solid contents (SSC). Majority of the techniques used to determine the maturity stages are destructive involving the removal of little quantity of fruit tissue especially for the measurement of SSC, total acidity and nutritional content. These techniques resulted in large amount of postharvest losses, inability to measure the whole batch and laborious. However, over the last decades several attempts have been made to develop optical techniques for monitoring quality indices of fruits via non-destructive approach. Laser light backscattering imaging (LLBI) system is one of the emerging optical techniques which is inexpensive and easy to use. Bananas samples at six ripening stages i.e. from ripening stage 2 to 7 and pear samples at different days after full bloom (dafb) were obtained from a ripening facility at Potsdam Bornim and Sachsenobst orchard Germany respectively. The samples were kept at 14 °C with 79 to 89 % relative humidity (RH). Laser light backscattering imaging (LLBI) with 5 laser diodes wavelengths in the visible and near infrared region i.e. 532, 660, 785, 830 and 1060 nm were employed to acquire the backscattering images of the samples and features were extracted from the backscattering images of both fruit using transform-based textural techniques viz: Wavelet transform, Gabor transform, Tamura texture and optical properties i.e. absorption and reduced scattering coefficients with Farrell’s diffusion theory. The reference measurements of index of chlorophyll, elasticity, firmness and SSC were measured with ΔA meter, texture analyzer, penetrometer and refractometer respectively immediately after backscattering images acquisition. The extracted features and optical properties at individual and combined wavelengths were used as an input into the prediction and classification models in the predictions and classification of quality attributes and maturity stages of both fruit. Two computational intelligence techniques, artificial neural network (ANN) and support vector machines (SVM) were used to build the prediction and classification models. Root mean square error of calibration (RMSEC), root mean square error of cross-validation (RMSECV), coefficient of determination (R2) and bias were used to evaluate the performance of the prediction models while overall classification accuracy was used to evaluate the classification models. The results showed that there was a very strong correlation between the absorption and reduced scattering coefficients with ripening stage of banana and pear development. The range values of absorption coefficient and reduced scattering coefficient of bananas at 532 nm were between 0.312 and 0.963, and 2.637 and 4.893 for ripening stages 2 to 7 respectively. The result shown a decreasing trend of optical properties with increasing wavelength. For pear at 532 nm, the range of absorption coefficient was between 0.033 and 0.308 while reduced scattering coefficient was between 3.160 and 6.741. For banana, analysis using ANN with visible wavelength region of 532, 660 and 785nm resulted in high R2 values ranging from 0.977 to 0.981 for the prediction of index of chlorophyll and 0.955 to 0.976 for elasticity; while near infrared region of 830 and 1060nm resulted in R2 range between 0.964 and 0.980 for SSC prediction when absorption and reduced scattering coefficients at individual and combined wavelengths were used. For the classification of banana into ripening stages 2 to 7, visible wavelength region using ANN gave the highest classification accuracy of 98.77% with combined wavelengths while the highest overall classification accuracy of 96.30 % was recorded at 830nm with SVM. For pear the highest R2 of 0.947 and 0.818 were obtained for firmness and SSC respectively using ANN while R2 values of 0.890 and 0.808 for firmness and SSC using SVM. For pear classification into different maturity stages, the highest classification accuracy of 90.42 % was achieved with both ANN and SVM. Similar results though lower were obtained when transform-based textural techniques were used for the prediction of banana and pear quality parameters and classification of banana and pear into different ripening and maturity stages. This study has shown that transform-based textural techniques and optical properties of banana and pears with ANN and SVM as prediction and classification models can be employed to predict the quality parameters and classify banana and pears into different ripening and maturity stages non-destructively.

Preview Text FK 2017 65 - IR.pdf Download (2MB) | Preview

Actions (login required)

View Item