abstract |
A method of object recognition based on salient objects in an image, the method comprising: a training process for building a classification database including first feature vectors describing a plurality of objects; and a recognition process , which includes: inputting a picture containing an object into a deep convolutional neural network, dividing the picture into M*M grids, predicting N candidate frames for each grid, and obtaining the probability of an object in each candidate frame ; when the probability of an object is greater than or equal to a predetermined threshold, select the candidate frame as the first effective candidate frame; input the image of the first effective candidate frame into the classification neural network to obtain the second feature vector; and based on the first The second eigenvector, the first eigenvector and the classification database execute the k-nearest neighbor classification algorithm (KNN) to identify the category of the object. |