Fruit Cluster Recognition and Picking Sequence Planning Based on Selective Attention

Abstract:

To improve algorithm versatility and picking efficiency, recognizing more kinds of fruits or vegetables and planning picking sequence for fruit clusters in visual field have become hotspots and trends of harvesting robot research. Primates can find objects and shift attention through visual selective attention mechanism, which has similarities with harvesting robots to recognize targets and to plan picking sequence. Hence, by adopting the idea and technology of bionics, a new visual selective attentionbased method was proposed for fruit cluster recognition and picking sequence planning. According to Itti visual attention computational model, this algorithm changed computing process for color feature maps, and improved feature integration method for building color conspicuity map and saliency map by introducing priori knowledge about fruits and vegetables. Using the reference of the biological neural network competition mechanism called WinnerTakeAll and artificial expertise from professional picking operators, distance, area and saliency were adopted to design the weighted preferential picking sequence strategy for fruit clusters. The experimental results showed that the proposed method in this paper achieved recognition of more than 23 kinds of familiar fruits and vegetables. Recognition correctness was higher than 93.36%. In addition, the results of picking sequence planning showed that its planning way was consistent with artificial expertise from manual picking operation.