Research On Information Fusion Technology In Grain Storage Safety Based On Deep Learning

As the world’s largest producer and consumer of grain,the post-production storage safety of grain in China is significant.Since grain storage is highly susceptible to environmental changes such as temperature and humidity,reasonable use of grain storage information and accurate judgment in the storage process is an effective way to ensure the safety of grain quality.In the face of complex grain analysis problems,it relies mainly on manual experience judgment and the need for a more efficient and stable method.The information fusion technology can fully explore the correlation and potential value between the information and provide adequate support for intelligent decision-making of grain storage to obtain safety assessment.Therefore,this paper is oriented to grain information fusion anomaly detection background and studies the multi-source information fusion technology based on deep learning.The deep learning fusion technology is used to conduct a comprehensive analysis of multiple grain storage information and assess the safety condition of grain storage.In this paper,we firstly construct a grain situation anomaly dataset to make up for the lack of anomaly information in the grain situation public dataset;secondly,we propose a two-level fusion model of grain situation information based on deep learning to fuse multiple grain storage information for decision making and finally verify the reliability and accuracy of the algorithm in grain situation anomaly detection through experiments.The main research contents are as follows:(1)Construction of grain situation anomaly dataset.To solve the problem of missing abnormal data sets of the grain storage environment,this paper collects abnormal condition data sets conforming to grain condition indicators by simulating the grain bin environment,deploying data dimension nodes such as temperature,humidity,and moisture,and making external adjustments.Subsequently,to reduce redundant and faulty data,optimized preprocessing,such as noise addition and reconstruction,is performed by a noise reduction encoder.In addition,the safety condition level of grain storage is classified according to the Food Security Storage Technical Index Evaluation System.This is the model training classification standard to compensate for the lack of abnormal condition data set for grain storage.(2)A two-level fusion model of grain storage safety information based on AWM-CNN is proposed.This paper adopts a two-level fusion approach of homogeneous and heterogeneous data fusion to ensure that the grain anomaly detection can accurately and effectively judge the grain storage safety condition.Firstly,starting from mining homogeneous feature information,the adaptive weighting method fuses the homogeneous information to reduce the complexity of feature fusion.Then the convolutional neural network model is improved to combine heterogeneous information by proposing a multi-layer branching convolutional neural network structure with feature extraction branches,deep feature extraction branches,and decision branches to strengthen the decision fusion accuracy.Finally,the experimental results show that the two-level fusion accuracy of the model reaches93%,which has a good effect on the abnormal classification decision of grain storage safety conditions.(3)The MAF-Transformer-based fusion model of grain storage safety information is proposed.To adapt to the complexity of grain situation changes and the diversity of environmental differences in each region,a multi-headed self-attention mechanism is used as the entry point to improve the feature extraction capability by fusing the output results of each self-attention module at the feature level.The Transformer model is then enhanced to improve the stability of model training and decision efficiency by designing the adversarial training module and the fuzzy cognitive module to generate malicious samples and learn the information association between multiple heterogeneous data sources.In addition,by adding focus scores as input sequences,the weight coefficients of crucial information of different grain situations are iteratively calculated to improve the accuracy of feature fusion matching data sets during model training so that the model can be adapted to different grain bin grain situations.Finally,the experimental results show that the model’s accuracy is 94%.It has better robustness and abnormal decision evaluation ability in grain storage safety information fusion to guarantee grain storage quality safety better.