Research And Implementation Of Plant Images Recognition Methods

Image recognition technologies are widely used in traditional manufacturing,secu-rity and the Internet industry,which are mostly mature.However,there are many gaps in biology,medicine,and food that need to be filled,mainly due to the fact that current identification methods are often directed to "large" category recognition such as cats and dogs,humans and vehicles,animals and plants other than the fine-grained image recognition,such as the identification of chrysanthemum indicum and chrysanthemum.We focus ourselves on the methods of plant image recognition which is a kind of fine-grained image recognition in this paper.The main works are as follows:1.Based on the basic descriptors of the image,we focused ourselves on the Bag of Visual Word and Fisher Vector feature coding method,we propose a multi-feature Fusion image recognition scheme which is based on the Fisher Vector,and integrates multiple features.Experiments show that the Fisher Vector-based feature coding scheme has a better effect in the application of the plant image recognition.2.Research on the deep learning methods' effect on the recognition of fine-grained images.Firstly,the contrastive experiments are set up to choose the primary mo-del,we delve into the effects of different training modes and convolution neural network depth on plant image recognition.Secondly,we propose a key region proposal method which is based on Selective Search algorithm;Finally,the paper proposes the VGGNet16-based plant image recognition model for key regions,and compares the experimental results with the public dataset and the self-built practical image set.3.We construct a plant image database.The database can be divided into two parts.The first part is the common plant public datasets,which are used for the horizontal comparisons of the methods.The other part is the self-built practical plant image dataset,which is used to verify the practicability of our methods.All experiments are carried out on these data sets.4.Finally,we design and implement a medicinal plant image recognition system.The system is based on the preferable methods we proposed in this paper.Besides,the user feedback mechanism of intermediate result and final result is designed to improve the system image recognition accuracy.