Research On Super-resolution Algorithm For Image Acquisition By Agricultural UAV

Agricultural UAVs have gained significant attention in agricultural research due to their flexibility,short cycle time,low cost,and ease of operation.The clarity of UAV acquisition images is crucial in collecting and monitoring agricultural production information.Aiming at the local blurring and redundant noise of agricultural UAV acquisition images,etc.,this paper proposes the super resolution algorithm for agricultural UAV acquisition images(AUSRGAN)to solve these problems.The main research work of this paper is as follows:(1)First,the image data obtained from agricultural UAVs were analyzed to create a suitable superresolution image dataset.Following this,a comparison of various super-resolution algorithms was conducted to weigh their respective benefits and drawbacks.Ultimately,a generative adversarial neural network-based algorithm was chosen as the optimal method for the reconstruction of the agricultural UAV acquisition images.(2)In this subject paper,we delve deeper into the generative adversarial neural network-based superresolution algorithm and explore ways to enhance its performance.First,we remove the batch normalization layer(BN layer)from the generative network part,which improves reconstruction efficiency.Additionally,we introduce an attention mechanism module to address issues related to incomplete image information extraction and corrupted image representations after reconstruction.(3)In addition,this paper presents a framework for local discriminative learning that can differentiate between pseudo-texture and real details produced by the generative adversarial neural network.This framework creates an artifact mapping map that helps standardize and stabilize the model training process,ultimately addressing the issues of artifacts and loss of details that arise when reconstructing detail-rich agricultural images.And an agricultural UAV acquisition image super-resolution(AUSRGAN)algorithm based on generative adversarial neural network is proposed.(4)Finally,the proposed AUSRGAN algorithm is ablated on the constructed dataset and compared with the typical super-resolution algorithm for experiments.The results show that the super-resolution reconstruction of ×2,×4 and ×8 magnification factors are performed under the same training conditions,and the AUSRGAN algorithm has better improvement in the objective evaluation indexes of peak signal-tonoise ratio,average value and structural similarity.At the same time,it has richer detail construction in subjective evaluation and artifacts are greatly reduced,thus achieving a comprehensive improvement in objective evaluation indexes and reconstructed visual effects.