Self-Configuring CVS to Discriminate Rocket Leaves According to Cultivation Practices and to Correctly Attribute Visual Quality Level

Abstract

Computer Vision Systems (CVS) represent a contactless and non-destructive tool to evaluate and monitor the quality of fruits and vegetables. This research paper proposes an innovative CVS, using a Random Forest model to automatically select the relevant features for classification, thereby avoiding their choice through a cumbersome and error-prone work of human designers. Moreover, three color correction techniques were evaluated and compared, in terms of classification performance to identify the best solution to provide consistent color measurements. The proposed CVS was applied to fresh-cut rocket, produced under greenhouse soilless cultivation conditions differing for the irrigation management strategy and the fertilization level. The first aim of this study was to objectively estimate the quality levels (QL) occurring during storage. The second aim was to non-destructively, and in a contactless manner, identify the cultivation approach using the digital images of the obtained product. The proposed CVS achieved an accuracy of about 95% in QL assessment and about 65–70% in the discrimination of the cultivation approach.