Image Segmentation of Bananas in a Crate Using a Multiple Threshold Method

In this paper, an automatic segmentation algorithm based on the threshold technique was designed for the image segmentation of bananas in a crate. The pixels of banana and background were selected discretely in the manually segmented regions, and subsequently, nine color features were extracted from RGB, HSV and CIE L*a*b* color space. Three thresholds, which were determined by statistical analysis of B, L* and b* color channel, were used for the design of the automatic segmentation algorithm. The intuitive comparison was applied for the qualitative assessment of segmentation performance. The results showed that the outlines of manually segmented and automatically segmented regions were almost similar to each other. The quantitative evaluation of segmentation performance was carried out by area ratio, and the results demonstrated that the average area ratio of 10 tested samples was more than 80%. These two tested results indicated that the performance of this automatic segmentation algorithm might be acceptable, and this methodology could provide a key technology to realize the real-time quality monitoring and control system in banana ripening rooms. Practical Applications An automatic image segmentation developed in the current study provides the potential of segmenting the bananas in a crate with the plastic bag. The algorithm is designed on the basis of a multiple threshold approach, thus allowing the high segmentation speed and accuracy. The obtained results demonstrated that the proposed multiple threshold-based segmentation algorithm could be used for achieving in situ quality monitoring and control system in banana ripening rooms.