Research And Implementation Of The Algorithm Of Flower Recognition Based On Machine Learning

Nowadays,with the rapid development of AI,the machine learning method combined with the image processing technology has been applied to all aspects of daily life.This thesis mainly uses the technology of machine learning to research the flower recognition and apply it: Use the superpixels-based method to realize the flower segmentation,and use the traditional feature fusion or convolutional neural network to realize the flower classification.The segmentation method based on superpixels can be more uniform and compact super-pixel segmentation,segmentation better,the traditional feature fusion method is relatively simple and easy to implement,the features extracted by convolution neural network can be a good performance of the internal images of flower images,with more excellent classification ability.In this thesis,a complete flower recognition algorithm is constructed based on the segmentation and classification algorithm,and experiments on flower datasets are performed to obtain a good accuracy of recognition.The main contents of this thesis include:(1)The SLIC flower image segmentation algorithm based on superpixel is studied,and the algorithm is improved and optimized.The input parameters of the algorithm are improved so that the number of pre-divided superpixels does not need to be artificially set before the segmentation,which simplifies the algorithm steps.Optimized the color conversion process,avoiding the huge floating-point calculation of RGB and Lab color space conversion,accelerated the calculation process of the algorithm.Optimized the distance measurement standard,the floating-point calculation is transformed into the plastic calculation,which completely avoids the huge calculation amount caused by the square root and the floating-point operation.(2)The classification of flower images based on feature fusion and convolutional neural network are studied.Based on the feature fusion method,the algorithm of LLC feature coding is emphatically studied.Then the multi-core learning method based on LLC,HOG and LBP is used to fuse different features.Finally,the results are compared by SVM classifier.Based on the convolutional neural network,the network model trained by AlexNet on the Image Net database is used.And use the migratory learning method to extract the features of flower images for classification.On this basis,we add the flower database to continue iterative training to fine-tune the network,get a new model suitable for flower feature extraction,and then use the new model to extract flower image features,and finally classify by SVM classifier and analysis the results.(3)Design the client and server framework of flower identification system,and combine the SLIC superpixels segmentation algorithm and feature fusion algorithm to realize the client of the identification system,and introduce the design and implementation process of the client frame in detail,and achieve the client to server data communication.The SLIC is combined with the convolutional neural network to realize the server of the identification system,and the server-side network model training and online identification process are described in detail.Finally,the system has been tested and analyzed in the self-built database,which meets the requirements of practical application both in real-time and accuracy.