Research On Bamboo Image Classification Algorithms Based On Convolutional Neural Network

As a kind of important renewable resources,bamboo has the advantage of fast growth,short cycle and high yield,and more and more attention has been paid to its ecological and economic benefits.However,due to the complex environment and similar appearance of bamboo,the number of species has reached more than 1200,and it is timeconsuming and laborious to rely only on manual classification.How to realize the automatic classification of bamboo species has been widely concerned by bamboo experts.In recent years,with the continuous development of deep learning in the field of image classification and recognition,it has features such as autonomous learning target feature ability,multiple convolution kernel and weight sharing,which provides the possibility for realizing automatic classification of bamboo species.Therefore,this paper focuses on three problems of bamboo image data collected under natural environment by using convolutional neural network: establishing a bamboo classification model,optimizesing the established classification model in combination with the saliency map,and transplanting the optimized model to the mobile phone to realize bamboo automatic classification and identification.The main research work of this paper is as follows:(1)The bamboo classification model based on convolutional neural network is compared.The common convolution neural network VGG16,Inception V3,Res Net50 and Dense Net121 are trained and tested on the original bamboo species dataset.The experimental result shows that the accuracy of VGG16,Inception V3,Res Net50 and Dense Net121 is respectively 71.05%,78.27%,83.29% and 87.31%,so the convolutional neural network model Dense Net121 with high accuracy is selected as the basic model of original bamboo classification.(2)The bamboo classification method by combining the significance detection algorithm RC and the convolution neural network model Dense Net121 is studied.This method use significant detection algorithm RC obtain saliency map,and through the Otsu to intercept bamboo significant region.Then the bamboo significant region images are normalized in size,and Dense Net121 model is used for bamboo classification training and testing.The experimental results showed that this method can reduce the interference of image background,effectively improve the bamboo classification effect,and further improve the bamboo classification accuracy.(3)The bamboo classification system based on Android system was developed.In this paper,the bamboo classification model of combining the bamboo significant region and Dense Net121 is selected and applied.The bamboo classification system was designed and developed on the Android Studio platform through Java language programming to realize the fast and convenient bamboo image classification.