Design And Implementation Of Experimental Data Collection And Evaluation System For Smart Lab

As an important part of talent training,experiment teaching is an important teaching means to help students digest and understand the theoretical knowledge,cultivate practical ability and innovative practical ability.However,under the current traditional mode of experiment teaching,students operate,observe and monitor a variety of experimental apparatus to record experimental data,which is prone to errors and inefficiency,leading to widespread plagiarism and fraud of experimental data to meet the requirements of experiment teaching.Meanwhile,teachers need to spend a lot of time for the assessment of students’ experimental results,and the evaluation process is vulnerable to the influence of subjective judgment and non-uniqueness of answers,which affects the assessment results and leads to inefficiency of experiment teaching.This paper improves the traditional experiment teaching mode,eliminates experimental data fraud,reduces the burden of teachers,and improves the efficiency of experiment teaching to achieve experimental data collection and evaluation.Supported by the information project of Guangxi Department of Industry and Information Technology "Development of Intelligence Experiment Teaching Platform and Curriculum Based on Artificial Intelligence",the following research works have been carried out :(1)In view of the existing problems with traditional experiment teaching,such as observation and monitoring of experimental apparatus to record experimental data,which are prone to errors and low efficiency,and students’ experimental data fraud,the experimental data collection based on instrument interaction is realized from the perspective of digitized experimental data.The system first consider the number corresponding between students and the experimental apparatus,through the students’ experiment data directly associated with experimental instrument to solve the problem of the experimental data of fraud from the sources,further,the experimental data obtained through the experimental data collected from inside the instrument,experimental apparatus to reduce the artificial operation and observation of the way to get data lead to errors and inefficiency.Finally,considering the need to use a variety of experimental apparatus in the circuit experiment,through the slave computer to achieve the unified management four kinds of experimental measuring apparatus,which oscilloscope,multimeter,signal generator and power.Meanwhile,the real-time test results show that the above four experimental apparatus can complete the experimental data collection within 6s,615 ms,700ms and 990 ms,respectively,because the student experiments are time-limited.(2)Due to the variety of experiments and the serious shortage of data samples in the current experiment teaching,it is difficult to use the method of traditional deep learning to automatically assess the image data onto the experimental report.In this paper,based on the idea of image classification,the experimental images are divided according to the response criteria to form a data set covering different segments,which provides data support for the implementation of intelligent classification evaluation in the next step.To select the appropriate algorithm,and further studied the existing evaluation methods as well as the deep learning related to the work principles and the key points,choose to use relation network based on few-shot learning as intelligent classification evaluation model,and in view of the network in the image classification evaluation are analyzed,and the deficiency in the corresponding improvement,A small amples Cclassification assessment method of experimental images based on improved relation network is proposed.Experiment results show that the proposed model improves the classification accuracy by 4.5%,1.91% on miniImage Net data set,and 1.54%,1.03% on CUB data set under the conditions of 5-way 1-shot and 5-way 5-shot.Further to verify the validity and feasibility of the model,the proposed model achieves 61.01% and 68.91% classification assessment accuracy under the conditions of 5-way 1-shot and 5-way 5-shot based on the experimental data of the intelligent laboratory of Guilin University of Electronic Science and Technology.