Recognition of citrus fruit and planning the robotic picking sequence in orchards

Abstract

To improve the operational efficiency of and to prevent possible collision damage in the picking of citruses by robots in densely planted complex orchards, this study proposes an algorithm based on YOLOv5 for recognizing citruses and planning a picking sequence. First, the convolutional block attention module (CBAM) is embedded into YOLOv5 to improve the feature extraction capability of the network and mitigate missed detection of occluded targets and small targets. Simultaneously, the bounding loss function is optimized to improve the positioning accuracy of the bounding box. This combined model is used to recognize and localize citruses. Then, a three-dimensional model of citrus fruit was established, and an adaptive pheromone concentration updating mechanism was introduced on the basis of the ant colony algorithm to dynamically judge the picking order of citruses and determine the optimal picking sequence. We show the quantitative and qualitative results of our method in comparison to previous methods. In tests, the F1-score of this method for citrus in a densely planted environment was 92.41%, which is a 2.81% improvement compared to YOLOv5. Compared to stochastic planning, the proposed method can plan the picking sequence of citruses in the field of view in advance and shorten the picking path. In addition, extensive sequence planning experiments on other fruits validate the superior of the proposed method. Therefore, the method in this paper may provide new solutions for citrus anti-collision picking and orchard yield forecasting and new ideas for the intelligent fruit industry.

A marigold corolla detection model based on the improved YOLOv7 lightweight

“…In recent years, with the rapid development of deep learning and the continuous open source of algorithms such as Faster R-CNN [3] and YOLO series [4][5] [6], the application of deep learning in the agricultural field has become more and more common. It is widely used on agricultural products such as citrus fruit [7], cherries [8], sugarcane [9], small wheat spikes [10], tomato [11], cotton [12], etc. Qi et al [13] proposed a highly fused and lightweight deep learning architecture for tea chrysanthemum detection based on YOLO (TC-YOLO), using CSPDenseNet and CSPRes-NeXt as the backbone network, combining the recursive feature pyramid (RFP) multiscale fusion reflow structure and the Atrous Spatial Pyramid Pool (ASPP) module with cavity convolution, which can achieve an average accuracy of 92.49% on their own tea chrysanthemum dataset.…”