Sorting recognition of taproots and stem bases of Panax notoginseng using multi-objective visual tracking algorithm

Abstract

For traditional Chinese medicine Panax notoginseng, it is necessary to manually sort or mechanically separate taproots and stem bases before pharmaceutical production. However, the taproot and the stem base are very similar in morphological structure and surface texture, which makes it difficult to be effectively distinguished on the sorting conveyor belt with similar color. Aiming at this problem, this paper proposed a sorting recognition algorithm of taproots and stem bases of Panax notoginseng based on multi-object visual detection and real-time tracking: The sorting of Panax notoginseng under the background of conveyor belt requires that at least one of the taproot or the stem base can be detected. Therefore, based on the detection algorithm, we have connected a multi-object visual tracking algorithm in series. Through the tracking algorithm, we can aggregate the characteristics of the same target from different angles in different time domains, and expand the detection at time points to the detection in continuous time domain, to strengthen the accuracy of sorting recognition. The experimental results on the actual photographed Panax notoginseng dataset in the test site show that the sorting recognition algorithm of taproots and stem bases based on multi-object tracking can achieve the optimal average MOTA of 78.06% at an average speed of 34.3FPS. The method can detect and identify the stem base in the scene of sorting conveyor belt, which is obviously better than other state-of-the-art algorithms, and can provide accurate detection data for industrial intelligent sorting actuator.