Abstract: With the deepening of the concept of "unmanned", the functional requirements of home service robot are higher and higher. The function of quickly taking designated daily necessities and accurately delivering them has strong market value. However, high-precision detection algorithms have high hardware requirements. Therefore, an improved YOLOv5s deep learning algorithm (YOLOv5ss) is proposed to identify daily necessities. The internal convolution operation was used to extract the features of the image instead of convolution operation, and the prediction and evaluation system was optimized based on CDIoU method, which effectively reduced the model complexity and sped up the recognition speed. Through comparative experiments, we found that the YOLOv5ss network greatly reduces model complexity and improves recognition speed while ensuring accuracy.