Parking space number detection with multi‐branch convolution attention

With the increase of large shopping malls, there are many parking spaces in complex environments, which increases difficulty finding vehicles such environments. To upgrade consumer's experience, some car manufacturers have proposed detecting space numbers spaces. The detection number environments has problems as diversified background numbers, tilted direction and small scale. Since no scholar a high-performance method for problems, model based on multi-branch convolutional attention is presented. Firstly, using ResNet50 backbone network, structure aims to process fuse feature map through three parallel branches, enhance network represent ability information by attention, learn global features selectively strengthen containing helpful information, improve detect area. Secondly, high-level enhancement unit designed adjust channel channel, obtain more spatial correlation, reduce loss generation. data results dataset CCAG show that precision, recall, F-measure 84.8%, 84.6%, 84.7%, respectively, certain advantages detection.