Research On The Method Of Image Compliance Detection Oriented To Bionic Model

In last few years,the Internet has become an indispensable part of people's lives along with the rapid development of the Internet,big data,and artificial intelligence.At the same time,the self-media industry has grown stronger as well.Some non-compliant images are inevitably presented on the Internet.This not only pollutes the network environment,but also affects people's physical and mental health to a certain extent.Secondly,with the growth of the information age,deep learning has become a prominent aspect of this category.Generally speaking,deep learning consumes a lot of computing power and memory.For neural networks,the more sophisticated the neural network,the more accurate the results will be.This makes the neural network model larger after deep learning,and a larger network model will increase the burden of embedded devices.The speed of network model operation and monitoring and the size of the model are very important for embedded devices.Therefore,energy consumption has also become an indicator that people are paying attention to.This paper mainly studies the compliance detection of artificial intelligence images based on the bionic model,and has verified the effectiveness of the algorithm in this paper through experiments.The main tasks include the following aspects:(1)Aiming at the increasing proliferation of non-compliant pictures under the continuous development of data,a method based on YOLO(you only look once)is proposed to detect the target of compliant pictures.Traditional target detection algorithms cause a lot of loss of manpower and material resources,take a long time,and have poor accuracy.Compared with traditional target detection algorithms,the use of deep learning-based target detection algorithms has faster detection speed and higher accuracy.In order to meet the needs of the experiment,a compliant picture data set was established,the data set was converted into a picture format and the data was manually labeled as a VOC format adapted to the YOLO algorithm,and target detection was performed on this data set.By comparing the speed,accuracy and compliance image data set of the detection results on different target detection algorithms,it shows that the algorithm in this paper can be more useful for target detection on this data set.(2)In view of the large number and wide range of non-compliant images appearing in the network,the target detection model obtained through training and learning is often large in size.This article connects the thinking of bionics,and further uses pruning on the neural network to shrink the model.For embedded devices,an excessively large network model will consume a large amount of memory and reduce the function of the device.In the neural network,some parameters have little effect on the final result,and there is redundancy.Pruning is to cut off these excess parameters.After pruning,the network model becomes more portable,achieving the goal of network slimming,reducing memory,and increasing throughput.The experimental results can show that,compared with the original network model,the compressed model reduces the size and speeds up the detection rate while the accuracy is almost unchanged.