Research And Implementation Of Digital Instrument Recognition Based On Lightweight Deep Learning

Real-view simulation technology based on motion scenes refers to the technology where people create real motion scenes and engage in interactive movements through video projection indoors,in order to achieve immersive motion.As a widely used sports device,treadmills can combine real-life simulation technology with treadmills to achieve real-life simulation of treadmills.This technology mainly consists of three parts: the application end that obtains the speed of the treadmill,the server end that coordinates the playback speed of the live video,and the display end that plays the live video.The acquisition of treadmill speed is the foundation and key to technological implementation.However,existing methods for obtaining treadmill speed have problems such as limited access,inaccurate acquisition,and increased implementation costs.Therefore,it is necessary to conduct research on obtaining treadmill speed.In response to the above issues,this article proposes to obtain the speed of a treadmill by real-time recognition of its instruments,and conducts research and analysis on the recognition of treadmill digital instruments.A digital instrument recognition algorithm based on lightweight deep learning is proposed and implemented,which mainly consists of three parts: digital display area positioning,character segmentation,and digital recognition.This article uses edge detection,Hough transform,and K-means clustering algorithms to locate the digital display area of the treadmill digital instrument.The vertical projection method is used to segment the numbers in the digital display area image into a single digital image,and a lightweight deep learning algorithm is used for digital image recognition.Finally,the identified numbers are recombined to obtain the speed value of the treadmill digital instrument.This article constructs a new lightweight deep learning algorithm inspired by the lightweight deep learning algorithm Mobile Netv2.Firstly,the size of the image input model is adjusted,and then the inverted residual structure is used to extract the feature information of the image,ensuring that the model has high accuracy with low computational and parameter complexity;Secondly,a lightweight channel attention module was added to the constructed model,which was added after deep separable convolution to extract important information from the feature layer and improve the overall performance of the model.The experiment shows that the lightweight deep learning algorithm proposed in this paper has lower computational complexity and can accurately recognize digital images in real-time.Finally,the image processing algorithm and trained digital recognition model were deployed to the Android application,and the recognition process was designed in detail according to the actual requirements of the system.Real time acquisition of preview video frames and frame image recognition were achieved,and the dynamic display of recognition results was also completed.After testing,the digital instrument recognition system has performed well in all aspects and can achieve real-time and accurate recognition of treadmill digital instruments.