Detection of Beer Bottle Caps with Defects Based on Improved ShuffleNet V2

Beer bottle caps are produced in large quantities. In addition, beer bottle caps with defects not only affect the customer's experience, but also may not protect the beer effectively. Therefore, it is important to detect defects in beer bottle caps quickly, efficiently and accurately. Deploying convolutional neural network (CNN) on field programmable gate array (FPGA) can achieve the above objectives. However, CNN generally has a large number of parameters and high complexity, in addition, CNN operates differently among different hardware devices. In order to solve the problems, this paper improves the ShuffleN et V2 so that the model is suitable to be deployed on FPGA, and runs fast on FPGA. The improved methods are shown below: firstly, we reduce the number of output channels in the Conv5 module. After that, we replace the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1\times 1$</tex> ordinary convolutions of the right branch in the ShuffleNet V2 Unit1 and Unit2 with the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1\times 1$</tex> group convolutions, and add the channel shuffle after the first group convolution. Finally, we replace the convolution operations of the left branch in the ShuffleN et V2 U nit2 with the max pooling. Experiments show that our proposed model not only reduces the number of parameters and the complexity of the model, but also improves the accuracy and the f1-score of the model.