Multi-scale selection pyramid networks for small-sample target detection algorithms
Hao Peng, Xiaoming Li · 2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI) · 2021
Target detection is to detect the specified target in the image. This technology has been widely used in automatic driving, face recognition, and other fields and has become a significant research hotspot in the field of computer vision at home and abroad. Traditional target detection often requires many annotated data sets, so detecting targets with only a small number of annotated samples is challenging. We propose a multi-scale selection pyramid algorithm for small sample target detection to address this problem so that detection no longer relies on large-scale labeled datasets. First, we designed a multi-scale selection pyramid network for small sample target detection, consisting of three components: Contextual Layer Attention, Feature Scale Enhancement, and Feature Scale Selection module. After the RPN network generates the RoI features, the max-pooling and average-pooling are used to extract the features and then merge them. The model's sensitivity to the new class of parameters was improved with the stability of the sample parameters. In addition, the orthogonal mapping loss function is used to constrain the features before the classification layer, which can reasonably measure the similarity between features even in the case of a small number of samples.