Research On Face Mask Occlusion Recognition Algorithm Based On Capsule Network

Facial detection and recognition has always been a hot topic in the field of computer vision.Especially in recent years,due to the impact of the COVID-19 pandemic,wearing masks has become the norm.As most of the facial area is covered,it has posed great difficulties to facial recognition.Traditional facial occlusion detection algorithms are mainly based on convolutional neural networks(CNN),but they still have problems such as low recognition accuracy and different sensitivity to different types of occlusion.Capsule networks were proposed to address some of the deficiencies of traditional CNN and have demonstrated their superiority in image classification.However,there have been very few studies on capsule networks in identity verification for faces with occlusion.Therefore,this article proposes an Improve Cap algorithm for facial occlusion recognition.The algorithm is based on the capsule network with improvements and attention mechanisms to improve its accuracy.Firstly,this article analyzes the advantages and disadvantages of CNN networks in facial occlusion recognition in detail.CNN networks have good feature extraction capabilities,which can effectively extract and learn local features of images,thereby detecting occlusion.However,since CNN networks perform convolutional calculations based on fixed convolutional kernels,they have certain limitations in processing different types of occlusion.To solve this problem,this article proposes a facial occlusion recognition algorithm based on capsule networks.By introducing capsule layers and using vectors to describe features,capsule networks achieve more accurate feature learning and extraction,thereby effectively improving occlusion detection accuracy.Secondly,to select a more suitable capsule network for facial occlusion recognition tasks,this article trained two generations of capsule networks on its self-made dataset.At the same time,to reduce overfitting during capsule network training and improve its generalization ability and performance,this article applied Dropout operations to the output of each capsule.By randomly setting some outputs to zero,the network is prevented from relying too much on certain capsules,and is forced to learn more robust features.Finally,in response to the insufficient expression ability of the capsule network for face occlusion recognition,this paper introduces a Convolutional Block Attention Module(CBAM)and compares it with other attention modules suitable for face classification tasks.Experimental results show that the proposed Improve Cap algorithm demonstrates high recognition rates in face occlusion recognition under complex environments,and has certain practical application value.