Recognition System Of Agricultural Machinery Operation Test Environment Based On Deep Learning

Operation test of agricultural machinery is an important link to evaluate and check the overall performance,reliability and quality of agricultural machinery,and is the key to improve the speed of research and development of agricultural machinery and product quality.According to the relevant provisions of the reliability test standard in the agricultural machinery identification and extension outline,the environment of the agricultural machinery operation test is clearly stipulated.The agricultural machinery undergoing the test must be operated in accordance with the prescribed scenes.Only in this way can the operation data generated during the agricultural machinery operation test be guaranteed to be authentic and reliable.Therefore,it is an important means of reliability test to identify the operating environment type of agricultural machinery.The traditional method to judge the position of agricultural machinery according to the location information has the disadvantage of wrong judgment because of the problem of signal blocking;Using images to identify working environment may also cause problems such as image occlusion and illumination,which will also lead to the decline of recognition accuracy and poor effect.Therefore,it is not only of great significance in theory,but also of great application value in practice to study a high-precision and high-robustness identification system for agricultural machinery operating test environment.The main work of this paper includes the following aspects:(1)For a single image source,an improved residual network(Res Net18)algorithm for image scene classification is proposed.After 4 categories of AID data set were selected for data enhancement,the three structural models were compared and analyzed to classify the data set and compare the accuracy.Finally,Res Net18 with the best effect was selected as the basic model for improvement.In terms of model structure,the attention mechanism is added,and Silu activation function is used instead of Relu activation function.In terms of model training strategies,label smoothing and cosine annealing algorithms are used.The experimental results show that the accuracy of the improved residual network(Res Net18)is improved,which proves the superiority of the model proposed in this paper.(2)Pavement type recognition based on EEMD-SVM is proposed for single vibration signal source.Firstly,the response of agricultural machinery under road excitation is analyzed,and the appropriate sensor signal object is selected for feature extraction.Because of the mode aliasing of EMD algorithm and the small sample characteristics of the actual collected signals,the set empirical mode decomposition(EEMD)is used to extract the features of the collected vibration signals,and the support vector machine(SVM)is used to classify the extracted feature vectors..Experimental results show that high average recognition rate is obtained by analyzing experimental data with this method.In addition,Compared with EMD algorithm combined with support vector machine,the result also shows a higher recognition rate,which proves the superiority of EEMD-SVM algorithm and achieves the expected effect.(3)Based on the improved residual network(Res Net18)and one-dimensional convolutional neural network(1DCNN),two kinds of signal sources were fused at feature level to identify the agricultural machinery operating environment type.In order to ensure the recognition with high precision and robustness,the corresponding relationship between image and vibration signal is established on the basis of the above data processing,and the accuracy rate of information fusion environment recognition is obtained.Compared with a single signal source,the fusion model has the characteristics of high robustness,and the recognition accuracy has reached the expected value,which solves the problem that in practical application,due to some special circumstances,the data acquisition of one channel is invalid and the agricultural machinery operating environment cannot be recognized.In addition,this paper designed a GUI interface,simulated a farm machinery operation test environment identification system,and realized semi-automatic environment type identification.