Quality assessment of fresh and fresh-cut fruit and vegetables is a complex process which commonly involves the use of analytical and destructive techniques that are time consuming, expensive and require polluting reagents and sophisticated equipment. Moreover, these methodologies are not suitable for in-line application on industrial lines, where, nowadays, speed, reliability, accuracy and sustainability are required. A computer vision system (CVS) provides a suitable alternative as contactless and non-destructive methodology to achieve a consistent quality assessment of fruit and vegetables, even on packaged products. The results presented are a part of a PhD project, carried out within the PRIN SUSTAINING LOW-IMPACT PRACTICES (LIP) IN HORTICULTURE THROUGH NON-DESTRUCTIVE (ND) APPROACH TO PROVIDE MORE INFORMATION ON FRESH PRODUCE HISTORY & QUALITY (SUS&LOW), aimed to develop and validate predictive models based on the use of CVS for the assessment of the quality level (QL) and the main quality parameters of fresh and packed rocket leaves. The proposed CVS is able to automatically select, without human intervention, the most relevant colour traits using the Random Forest as machine learning model. During first experiments, CVS was applied to fresh-cut rocket leaves, obtained by low-impact agricultural practices, to objectively assess its quality levels (QL) during the storage at 10 °C according to a 5 to 1 rating scale and to discriminate the fertilization levels and irrigation managements applied during the cultivation. Promising results showed an accuracy of 95% in the QLs assessment and of about 65-70% in the discrimination of the cultivation approach. Then, five experiments were conducted to validate the CVS in estimating internal quality traits (chlorophyll and ammonia content) related to the shelf-life loss of rocket leaves, even though the package. Similar performances were obtained on packaged (Pearson’s coefficient of 0.84 for chlorophyll and 0.91 for ammonia) and unpackaged products (0.86 for chlorophyll and 0.92 for ammonia). Finally, PLS models well forecasted the VQ of rocket leaves using as predictors the chlorophyll content obtained by destructive methods and by CVS on packaged and unpackaged products (R2v of 0.70, 0.77 and 0.80, respectively).
Computer vision system for non-destructively evaluating quality traits in fresh and packaged rocket leaves
Montesano Francesco F.;
2023-01-01
Abstract
Quality assessment of fresh and fresh-cut fruit and vegetables is a complex process which commonly involves the use of analytical and destructive techniques that are time consuming, expensive and require polluting reagents and sophisticated equipment. Moreover, these methodologies are not suitable for in-line application on industrial lines, where, nowadays, speed, reliability, accuracy and sustainability are required. A computer vision system (CVS) provides a suitable alternative as contactless and non-destructive methodology to achieve a consistent quality assessment of fruit and vegetables, even on packaged products. The results presented are a part of a PhD project, carried out within the PRIN SUSTAINING LOW-IMPACT PRACTICES (LIP) IN HORTICULTURE THROUGH NON-DESTRUCTIVE (ND) APPROACH TO PROVIDE MORE INFORMATION ON FRESH PRODUCE HISTORY & QUALITY (SUS&LOW), aimed to develop and validate predictive models based on the use of CVS for the assessment of the quality level (QL) and the main quality parameters of fresh and packed rocket leaves. The proposed CVS is able to automatically select, without human intervention, the most relevant colour traits using the Random Forest as machine learning model. During first experiments, CVS was applied to fresh-cut rocket leaves, obtained by low-impact agricultural practices, to objectively assess its quality levels (QL) during the storage at 10 °C according to a 5 to 1 rating scale and to discriminate the fertilization levels and irrigation managements applied during the cultivation. Promising results showed an accuracy of 95% in the QLs assessment and of about 65-70% in the discrimination of the cultivation approach. Then, five experiments were conducted to validate the CVS in estimating internal quality traits (chlorophyll and ammonia content) related to the shelf-life loss of rocket leaves, even though the package. Similar performances were obtained on packaged (Pearson’s coefficient of 0.84 for chlorophyll and 0.91 for ammonia) and unpackaged products (0.86 for chlorophyll and 0.92 for ammonia). Finally, PLS models well forecasted the VQ of rocket leaves using as predictors the chlorophyll content obtained by destructive methods and by CVS on packaged and unpackaged products (R2v of 0.70, 0.77 and 0.80, respectively).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
