Online hand-sketched engineering drawing neural network recognition

For the recognition of online hand-sketched engineering drawings, many concepts such as a sketch´s center of gravity, gravity radii and regularized gravity radii (RGR) are introduced and a hand-sketched neural network recognition approach is proposed. In this approach. the hand-sketched classifier is constructed by extracting the RGR of hand-sketched primitives as their features, by crossing the four primitives´ RGR values as the learning samples of a BP neural network, which is trained by using the adaptive learning algorithm of gradient descent plus momentum item. The experiments demonstrate that not only can the classifier recognize the hand-sketched primitives of arbitrary directions and positions but also its abilities of anti-noising and identification are very robust. Furthermore, it needn´t be retrained in the application. These methods are also significant for the recognition of scanning engineering drawings.