Target Detection Based on Taylor Expansion and Bilateral Symmetric Network

In the task of target detection, traditional detection algorithms are prone to weak generalization and low detection accuracy due to the small number of parameters and fixed parameter values. Conversely, CNN-based detection algorithms are more accurate, but cannot be developed on mobile due to model issues. In this paper, we refer to the gradient features of HOG and propose the Taly preprocessing method which can preprocess images using Taylor extensions and extract multi-order gradient features of the images. Then TaylorNet is designed under the bilateral symmetric network. Multi - gradient features contain rich edge feature information of images. Then, the gradient feature is fused through the bilateral symmetrical network structure to achieve the fusion of low resolution and high resolution, so as to realize the accurate positioning and detection of edge features, and finally achieve the accurate detection of the target. Through the training and testing of the dataset SBD and VOC2012, the comparison results show that compared with some SOTA algorithms, TaylorNet effectively reduces the size of the model while ensuring high accuracy, so it can also effectively implement mobile development.