结合主动光源和改进YOLOv5s模型的夜间柑橘检测方法

Detection method of citrus in nighttime environment combined with active light source and improved YOLOv5s model

【目的】解决夜间环境下遮挡和较小柑橘难以准确识别的问题,实现采摘机器人全天候智能化作业。【方法】提出一种结合主动光源的夜间柑橘识别方法。首先,通过分析主动光源下颜色特征不同的夜间柑橘图像,选择最佳的光源色并进行图像采集。然后,提出一种夜间柑橘检测模型BI-YOLOv5s,该模型采用双向特征金字塔网络(Bi-FPN)进行多尺度交叉连接和加权特征融合,提高对遮挡和较小果实的识别能力;引入Coordinate attention(CA)注意力机制模块,进一步加强对目标位置信息的提取;采用融入Transformer结构的C3TR模块,在减少计算量的同时更好地提取全局信息。【结果】本文提出的BI-YOLOv5s模型在测试集上的精准率、召回率、平均准确率分别为93.4%、92.2%和97.1%,相比YOLOv5s模型分别提升了3.2、1.5和2.3个百分点。在所采用的光源色环境下,模型对夜间柑橘识别的正确率为95.3%,相比白光环境下提高了10.4个百分点。【结论】本文提出的方法对夜间环境下遮挡和小目标柑橘的识别具有较高的准确性,可为夜间果蔬智能化采摘的视觉精准识别提供技术支持。

【Objective】To solve the problems of occlusion and difficult identification of small citrus in the nighttime environment and redlize the all-weather intelligent operation of picking robots.【Method】A nighttime citrus identification method combined with active light sources was proposed in this paper.Firstly,the best illuminant color was selected by analyzing nighttime citrus images with different color features under active light sources.Then,a nighttime citrus detection model named BI-YOLOv5s was proposed with bi-directional feature pyramid network(Bi-FPN)for multi-scale cross connection and weighted feature fusion to improve the detection performance of occlusion and small citruses.The coordinate attention(CA)module with attention mechanism was introduced to further strengthen the extraction of target location information.Meanwhile,the C3TR module integrated with a Transformer structure was adopted to reduce the computing amount and better extract global information.【Result】The precision, recall and average precision of the citrus detection using the BI-YOLOv5s on test set were 93.4%, 92.2% and 97.1%, respectively, with 3.2, 1.5 and 2.3 percent higher than the YOLOv5s, respectively. Moreover, the identification accuracy of the proposed model with an active light source for nighttime citrus was 95.3%, with 10.4 percent higher than the model in the white light environment. 【Conclusion】The proposed method in this paper has high accuracy for the identification of occlusion and small target citrus in the nighttime environment, and it can provide technical support for nighttime visual identification of intelligent picking of fruits and vegetables.