Based On Convolutional Neural Networks Compression For Agricultural Image

In recent years,the development of intelligent agriculture is also very rapid,agricultural image processing is also a necessary demand in intelligent agriculture.Convolutional neural network also shows its great advantages in image visual processing.In the field of agriculture,the application of neural network to the identification of diseases and insect pests of crops and the classification and classification of fruits has also achieved breakthrough research results.However,in the field of agricultural image compression,the performance of image encoding and decoding,imaging quality,etc.,still need to be improved,which has a lot of research space and research value.This paper will use the development of deep learning in the direction of computer vision to make further research on the compression of agricultural images.This paper based on the convolution neural network this emerging technology for the image compression agriculture research,aimed at the existing agricultural image compression codec performance is not high,using a kind of image compression based on convolutional neural network content weighted framework to implement agricultural image compression,the convolutional neural network is applied to the agricultural areas of image compression.This compression framework solves the problems caused by the application of convolutional neural network in image compression quantization and entropy rate estimation,and obtains a better compression effect than traditional image compression methods(such as JPEG,JPEG2000 and BPG).In this paper,a content weighted importance graph is introduced into the image compression structure,which can better guide the allocation of local bit rate in the compression process,control entropy coding according to the importance of local information,and reasonably remove the information that is relatively less important in people’s eyes.According to the experimental results,the network framework proposed in this paper has a greater improvement in performance than the traditional methods,and has been applied to the agricultural image field and obtained satisfactory visual effects.Then,the image quality of agricultural compressed image is not high,such as the low level of image monitoring equipment in greenhouse agriculture,which results in the poor image quality of agricultural image.Such as compression image quality can be further improved.In this paper,a residual dense network framework is used to realize the superresolution reconstruction of agricultural image compression by using convolutional neural network for the post-processing stage of agricultural image compression.In the reconstructed network framework,a variety of residual modes and feature fusion are used to enhance the learning of the input agricultural image features,among which the local intensive residual module and the global residual learning can make the network better trained and the detailed parts of the agricultural image can be better learned.According to the intelligent agricultural greenhouse monitoring graphics data sets to get the experimental results show that the proposed method can effectively enhance the quality of the image from agricultural to compress image quality has improved,image quality on the visual effect is also satisfactory,can more adapt to meet intelligent agricultural development under the background of image using demand for agriculture.