Complex Background Ground Based On Artificial Intelligence Research On Multi-Objective Extra Ction Technology

Deep learning is the main driving force for the development of artificial intelligence and plays a leading role in the field of intelligence,laying the foundation for the future development of intelligent battlefield operations.The extraction and detection of military targets is crucial for the mastery of military information.The complexity of modern combat environments is gradually increasing,so it is necessary to strike enemy targets more accurately.How to use deep learning to extract tank targets more accurately in complex backgrounds and improve the transformation and innovative development of deep learning in the military field has great application value in practical military strategies and situations.Firstly,in response to the issue of a small dataset,this article uses mobile phone images of tank models as the original dataset,and sends this dataset into the adversarial generation network to generate new tank images.In order to improve the performance of the generator and discriminator,generate clearer images,reset the training ratio of the generator and discriminator,and introduce different additional information into the generator and discriminator respectively.Secondly,when detecting tank targets on the battlefield,tank targets often have camouflage and are hidden in the jungle,which can be easily obstructed by trees,making it difficult to extract some target features,resulting in a decrease in target detection rate.To solve this problem,this article studied and improved the Faster R-CNN algorithm,introducing repulsion loss and increasing the distance between the prediction box and surrounding non-target boxes,Make the target prediction box appear more accurately on the tank target area,reduce the target missing detection rate,at the same time,fuse the backbone feature extraction network Res Net with the FPN network,reset the anchor proportion and size,optimize the candidate area,obtain more network Semantic information,extract more detailed features of the target,and improve the detection performance of the model.Afterwards,in response to the problems of high computational complexity and slow running speed in the detection network,the improved Faster R-CNN network was further optimized.Firstly,the importance of weights is evaluated by combining the absolute value of weights with the amount of weight variation.On this basis,in order to prune more useless parameters in the initial stage of network pruning,a progressive iterative pruning algorithm is introduced to obtain a more efficient sparse model.This method can effectively reduce the number of parameters in the network with minimal impact on network accuracy variation.Finally,verification and analysis were conducted on the three aspects of the above research,confirming that the improved algorithm in this paper has achieved certain improvements in accuracy(increased by 6.45%)and speed(accelerated by 1.58 times),with better performance and effectiveness.