Research And Implementation Of Flower Recognition Based On Convolutional Neural Networ

With the improvement of people’s lives and the level of technology,image processing and pattern recognition have made great progress.Flower category recognition,as a typical fine-grained image recognition task,is one of the most popular research topics in the field of computer vision and forestry informatization.With the increasing demand for flower recognition applications,deploying flower recognition algorithms on websites has become an urgent task.Traditional machine learning flower recognition algorithms are relatively mature,but traditional flower recognition methods use experts to manually select multiple features for classification,and face problems such as low accuracy of classification results,weak generalization ability,and long classification time.According to the limitations of traditional methods,this paper proposes a flower recognition algorithm based on deep learning.This thesis proposes an algorithm for flower recognition based on the improved Res Net18 network model.This article mainly starts the research work from the following aspects: 1.Feature extraction.In the feature extraction stage,the first layer of basic convolution in each residual block is replaced with a hollow convolution with an expansion rate of 2,and more detailed features of the flower picture are extracted to enhance the model’s ability to extract flower features.2.Introduce attention mechanism.By adding an improved channel attention mechanism after each residual block to obtain the network weights of local features,an improved Res Net18 network model is constructed.3.Model compression,using a shortcut pruning strategy based on residual network to reduce the amount of model parameters.The experimental results show that on the Oxford 102 Flowers Oxford flower dataset,the improved Res Net18 network model recognition accuracy can be as high as 97.78%,which is 3.11 percentage points higher than the model using only the convolution of holes,and 4.45 percentage points higher than the original model.The improved Res Net18 network improves the accuracy of flower recognition and enhances the generalization and fitting capabilities of the model.On the basis of the above work,this thesis uses Python language and Django framework to design a set of flower recognition system.The causes of image distortion are analyzed,and the perspective transformation matrix is used to process the image.The overall framework and functional design of the system are introduced in detail.The user can classify flowers only by clicking on the URL.