Report: Research on bet365 live slot machine Goods Detection Algorithm Based on Hybrid Attention Network
Report time:2022Year5month30Day afternoon3:00
Reporting location: Tencent Conference (Conference Number:902 589 286）
Reporter: Dr. Peng Jianbiao
Report Summary:
In recent years，Deep learning technology has shown extremely powerful performance in various image analysis tasks。In the field of bet365 live slot machine detection，Convolutional Neural Network (Convolutional Neural Networks, CNNs) is widely used in bet365 live slot machine currency、Detection of bet365 live slot machine medicines and bet365 live slot machine luxury handbags，And achieved certain results。However，Due to the following three issues，Accurate bet365 live slot machine detection remains challenging: (1) Fine-grained classification; (2) Category imbalance; (3) High imitation sample。In order to solve these problems，Proposed a hybrid attention network (HANet) to detect bet365 live slot machine goods.
atHANetA hybrid attention module is designed in。Use classic directlyCNNCompared to existing methods for bet365 live slot machine detectionHAThe module jointly uses channel attention units and spatial attention units to learn important bet365 live slot machine in channel and spatial dimensions。 HAModule can be easily integrated into ResNet In architecture to enhanceCNN’s discriminative representation ability，Thus helping the network discover the subtle differences between real and fake products。Also，Proposed an appraiser-guided loss for trainingHANet。Considering the factors of class imbalance and high imitation samples，The proposed loss gives a higher weight to the bet365 live slot machine class，At the same time, give higher weight to high imitation samples。The proposed loss introduces the knowledge of the appraiser，This makes HANet Not only can true and false samples be treated relatively fairly，And pay more attention to the learning of difficult samples。To evaluate the performance of our method，We built a well-benchmarked commodity dataset。on this dataset，Compared HANet、ResNet50 and the performance of state-of-the-art attention methods. The results show that,HANet Achieve better performance among all competitors.
About the speaker:Peng Jianbiao，Ph.D.’s main research interests include machine vision、Digital image processing and intelligent analysis。Mainly using digital image processing technology、Machine learning and deep learning technology，Combined with computer intelligence analysis，ImplementationAuthenticity of non-standard goodsDetection.
Contact: Peng JianbiaoContact number:13701346566Contact email:pengjianbiao@csu.edu.cn
