Software for classification of banana ripening stage using machine learning

Abstract

Abstract: Pattern recognition aims to classify some datasets into specific classes or clusters, having several applications in agriculture. The objectification of the process minimizes errors since it reduces subjectivity, allowing a fairer remuneration to the producer and standardized products to the consumer. Thus,this work aimed to develop an embedded system with artificial intelligence to determine the ripening stage of bananas (outputs) from the insertion of physical (i.e., fruit weight, texture and diameter), physicochemical (i.e.,pH, titratable acidity (TA), soluble solids (SS) and SS/TA ratio) and biochemical (i.e., total sugars, phenolic compounds, ascorbic acid,quantification of pigments in fruit peel and pulp and antioxidant activity by DPPH and FRAP methods) data (inputs). The bananas were harvested at each evaluated stage according to the Von Loesecke ripening scale, as follows:stage 2, totally green; stage 4, more yellow than green; stage 6, yellow; and stage 7, yellow with brown spots. Subsequently, they were selected and submitted to quality analysis. The data obtained were then mined and the attributes were selected using WEKA software. The classifier software was developed using MATLAB. The most relevant attributes selected in the Bayes Net classifier for the Cross-Validation method were: apical, central, basal and mean textures (between apical, median and basal textures), pH, soluble solids, phenolic compounds, antioxidant activities by the FRAP and DPPH methods, vitamin C, anthocyanins from the pulp, chlorophyll a content in the fruit peel and sugar, resulting in a mean F-measure of 97.0%.

Keywords

• Agriculture 4.0
• vegetable sorting
• data mining
• Musa spp.
• post-harvest