针对废旧塑料瓶在回收过程中不同颜色存在价值差异的情况，为解决颜色识别分选问题，提出一种基于深度学习的YOLOv8n改进模型废旧塑料瓶颜色高效识别方法。在颈部（neck）网络中引入加权双向特征金字塔网络（BiFPN）进行多尺度特征融合，提升模型的泛化能力；头部（head）网络的解耦头（decoupled-head）结构中两个分支均仅采用1个Conv2d模块，并在分支前端添加重参数化卷积——RepConv模块，减少计算量并提升训练精度；选用WIOU v3损失函数替换CIOU损失函数，提升模型的检测精度。通过对比实验可知，文章提出的模型优于传统目标检测模型。结果表明：文章提出的模型参数量较原模型减少了44.8%，计算量较原模型减少了34.6%，在50%交并比下的均值平均精度（mAP50）较原模型提升了2.7%，对废旧塑料瓶颜色进行识别时，每秒检测帧数（FPS）可达66，较原模型提高了40.4%，实现了对废旧塑料瓶颜色实时且准确的检测。

To address the value difference between different colors in the recycling process of waste plastic bottles, the study provides an efficient color recognition approach for waste plastic bottles based on a deep learning-upgraded YOLOv8n model to handle the problem of color recognition and sorting. Adding a Bidirectional Feature Pyramid Network (BiFPN) to the neck network for multi-scale feature fusion to improve the model's generalizability. The decoupled head structure of the head network uses only one Conv2d module for both branches, and a reparameterized convolution RepConv module is added at the front end of the branch to reduce computational complexity and improve training accuracy. Replace the CIOU loss function with the WIOU v3 loss function to improve the model's detection accuracy. Comparative trials demonstrate that the model suggested in the paper is superior to typical object detection models. The results showed that the model in the article had a 44.8% lower parameter count, a 34.6% lower computational complexity, and a 2.7% higher mean average precision at 50% IOU (mAP50) than the original model. When identifying the colors of waste plastic bottles, the frames per second (FPS) can reach 66, which is 40.4% faster than the original model. The colors of waste plastic bottles may now be detected in real time and with high accuracy.