Research And Application Of Lightweight Sketch Recognition Algorithm Based On Mobile Terminal

With the development of information age,Internet resources have been transformed from single text to rich multimedia content,especially images.Among them,sketch has attracted much more attention due to its immediacy and simplicity.But different from the objective reflection of natural images on things,sketch is the embodiment of the painter's subjective impression.Therefore,sketch has a higher diversity and the research of the computer's ability to recognize sketch has become one of the challenging topics in the field of image recognition.In addition,the popularity of intelligent touch-screen devices and the promotion of instant messaging software have largely changed people's communication methods.Since only a few lines used to draw sketches that can express things,people can communicate with each other anytime and anywhere.However,the current identification network parameters are large and cannot be directly embedded into the phone memory,so it cannot provide quick service on sketch recognition.Therefore,it is of great practical significance to study the lightweight sketch algorithm aiming at improving the recognition accuracy and speed.In this dissertation,a lightweight sketch recognition algorithm based on mobile terminal is proposed.Firstly,the depth separable convolution is introduced to reduce the number of network parameters,and the size and number of the convolution kernel are adjusted according to the sparse characteristics of the sketch.Then,two attention mechanisms of space and channel are applied to the sketch recognition field,and a sketch recognition network based on attention mechanism is designed in order to fully extract the sketch image features,In addition,the short-term memory network is used and a sketch recognition network based on strokes sequence is proposed considering the sequence features of sketch strokes.Experiments show that the sketch recognition network proposed in this dissertation can not only speed up the recognition,but also improve the accuracy.Finally,the sketch recognition system is implemented on mobile terminal.implementation of the above-mentioned lightweight sketch recognition model was successfully applied on mobile terminal,providing users with an application scenario of rapid sketch recognize.