| In today’s era of big data,with the rapid development of computer and Internet technology,the application scenarios of face recognition technology are becoming more and more extensive,and the technical branches of face recognition have also achieved different degrees of development and progress.Age and gender are important biological attributes of a face,so identifying the age and gender of a face image is also an important detection task of face attribute recognition,and age and gender detection based on face images can be applied to many scenarios,such as: advertising suitable for crowd placement,intelligent seat giving system for buses,intelligent physical examination classification of hospitals,etc.Thanks to the birth and transformation of convolutional neural networks,so far,the PC platform has emerged many classification models for face age and gender detection,and has a fairly good detection accuracy,but with the increasing frequency of this technology in life scenarios,the non-portability and high cost of PC devices are becoming more and more prominent,and mobile devices Although solving the problem of non-portability,but due to the huge amount of parameters and calculations of convolutional neural networks and complex network model structure,The chip computing power of mobile devices often does not have the ability to carry these convolutional neural networks and perform real-time calculations.In view of the above contradictions,this paper carries out the following work content,and proposes a scheme for face age and gender detection based on embedded device Raspberry Pi 4B.Firstly,by learning the classical convolutional neural network model in the image classification task,and proposing an improved network model based on VGG13,this paper realizes the reduction of the parameter amount and the improvement of the recognition accuracy of the network model by preprocessing the dataset and the appropriate increase of batch standardization layer and Dropout layer in the network structure,and reduces the number of parameters of the convolutional neural network by replacing the fully connected layer with the global average pooling layer.Experimental data show that the number of network model parameters is reduced by 92.9%,and based on the Adience face dataset,the accuracy of face gender detection and age detection is increased by 2.2%and 1.2%,respectively.In order to further break through the limitations of the traditional convolutional neural network model parameters and accuracy,this paper further turns its attention to lightweight convolutional neural networks,and proposes a Mobile Net V2 network model structure based on CBAM attention mechanism module through multi-faceted analysis of model size and detection accuracy,and shows through experiments based on Adience face dataset that the integration of CBAM attention mechanism improves the detection accuracy compared with the original Mobile Net V2.The gender detection rate and age detection rate increased by 1.8% and 1.5% respectively,while the amount of parameters increased was only about 4.9%,which was almost negligible,which was an effective way to improve the network model,and finally according to the idea of the control variable method,the improved convolutional neural network models Modified-VGG13 and Mobile Net V2-CBAM in this paper were compared with some common network models in the industry for a comprehensive comparison of parameter quantity and recognition accuracy.The results show that Mobile Net V2-CBAM is at the top of the two evaluation indicators,which is an ideal port for mobile deployment of neural network models.Finally,through the comprehensive consideration of the capabilities and cost control of the convolutional neural network model,this paper selects the embedded device Raspberry Pi 4B as the mobile platform of Mobile Net V2-CBAM,and after configuring the deep learning environment,converting the model format,integrating the model and real-time computing processing,Raspberry Pi 4B physical connection and a series of operations.Mobile Net V2-CBAM was successfully ported and deployed in the mobile device Raspberry Pi 4B,and through the test of the hardware system,the data showed that the system processed the picture frame rate of about 15 FPS,and the detection accuracy was also comparable to the detection accuracy of the PC side,which proved the feasibility of the application of the Raspberry Pi 4B-based face age and gender detection system. |
