Research on Classification and Detection System of Common Household Tools for Home Service Robot

In recent years, the application of vision processing algorithm based on deep network in robot and other mobile devices has become a research problem that attracts wide attention. In order to solve the problems of limited storage space, long prediction time, low algorithm performance and weak computing power of object detection on mobile devices such as home service robot, this paper designs a classification detection model of common household tools based on the lightweight convolution neural network MobileNetV2[1]. Firstly, MobileNetV2 is selected as the backbone network of feature extraction. By decomposing the standard convolution into deep convolution and pointwise convolution, the multi-scale prediction part is reserved, and the parameters are effectively reduced; then, the full connection layer network and Softmax classifier are used to realize the classification and recognition of common household tools. Compared with the common classification algorithms, the network algorithm has higher prediction accuracy, smaller network model and better performance, so it is better applied to the system platform of home service robot.