Recognition-based image segmentation of touching pairs of cooked shrimp (Penaeus Orientalis) using improved pruning algorithm for quality measurement

•Develop a classification model that can precisely recognize different shrimp types.•Improve the traditional pruning algorithm to separate touching shrimp efficiently.•Develop a combination method to inspect, segment and classify shrimp on-line.

Shrimp that touch each other can be slightly misleading in assessing shrimp quality using a machine vision system. To address this problem, a recognition-based segmentation technique called improved pruning method based watershed algorithm is presented. 352 individual shrimp was used to construct the classification model in the training stage, average hit ratio of 96.39% and average standard deviation of 0.0265 were achieved with three repetitions. In the testing stage, 247 touching shrimp were tested using the proposed algorithm under six touching scenarios. Results indicate that the proposed algorithm can successfully separate the touching shrimp with average hit ratio of 90.94% and average standard deviation of 0.065 on touching shrimp dataset with three repetitions. This outcome demonstrates that the proposed algorithm is effective and appears to be sufficiently robust to separate most of the touching scenarios with a few exceptions when shrimp are overlapped or have rough boundaries.