Image And Data Analysis Based On Artificial Intelligence In Industrial Internet

With the rapid development of the industrial Internet field and deep learning technology,deep learning technology has been integrated into all aspects of the industrial Internet field.This article takes the integration of industrial Internet and deep learning as the research entry point.Image recognition and target detection have always been hotspots in the combination of deep learning and industrial Internet,and they are also the basis for the integration of various technologies.The traditional industrial Internet encounters image recognition and target detection tasks with low efficiency and large errors,which are mainly reflected in: 1)The problem of noise in the feature image cannot be solved when recognizing the image;2)The feature extraction of the feature image is insufficient,which is easy to cause The main information and detailed information are chaotic;3)The network model has a single structure,which cannot cope with complex sample data,and the stability of the network model is poor;4)Target detection is prone to problems of missed detection and low recall rate,insufficient data pan-China capabilities,Too much overhead when multiple targets.This paper mainly proposes a combination of deep learning in image recognition and target detection in the traditional industrial Internet,and proposes a multi-dimensional branched convolutional neural network model and a target detection model based on deep learning.Firstly for the preprocessing of sample data,due to the complexity of the industrial Internet field,in view of the problem of image noise in the industrial Internet field,the metric multi-dimensional autoencoder pair composed of metric learning ideas,sparse autoencoders and denoising autoencoders is used.The data is preprocessed,and the autoencoder structure in the model is changed according to the specific conditions of the noise to remove the noise,and the sample data is filtered by the lightweight operation.Then start with "enhanced feature extraction" using Alex Net model,Group Normalization normalization algorithm,DenseNet and other technologies to build a network model,introduce the SENet sub-frame model to the network model,and appropriately adjust the structure of the network model according to the experimental results.Secondly,for the segmentation problem and multi-scale problem in the target detection field,the latest YOLO algorithm model is used to explore a variety of improvement methods,including adjusting the loss function,using the new non-maximum suppression value algorithm Soft?NMS to solve the non-maximum suppression value,Modify the file parameters to solve the reduction of the model's pan-China ability,change the network model structure to improve the model accuracy and training speed,and adjust the network model residual block to deal with complex situations.The experimental results prove that through the deep learning fusion of the Industrial Internet,this paper effectively solves the problems of high noise,feature deviation,single model,low recall rate,and high cost of sample data in image recognition and target detection of the Industrial Internet,making the Industrial Internet and Deep learning technologies are effectively combined.It is expected to play an important role in industrial Internet image recognition and target detection.