Study On Image Recognition And Classification Of Alien Invasive Plant Leaves

With the more frequent Global trade exchanges,the more opportunities for foreign plants to enter China.Alien plants will inevitably affect the natural environment of our country,and may even endanger people’s health.Rapid and accurate identification of alien invasive plants is of great significance to the protection of biodiversity and economic environment in China.In this paper,the extraction and application of leaf features and the establishment of recognition model of alien invasive plants were studied.This paper mainly analyzes the shape,texture and color of plant leaves,and determines 33 features to describe the properties of plant leaves.The shape features include dimensionless features and Hu invariant moment features determined by the geometric properties of leaves.The texture features are composed of gray level cooccurrence matrix features and Gabor texture features,while the color features are determined by color moments.A leaf recognition model of alien invasive plants based on leaf image feature analysis was proposed.This method is based on the BP neural network algorithm in machine learning.After preprocessing the extracted 33 kinds of blade features,they are sent into the model for training.The accuracy rate of identification on the self built leaf dataset of alien invasive plants was 90.63%.In this paper,a convolutional neural network recognition model based on the fusion of Gabor texture features and high-level semantic information of invasive plant leaves is proposed.The parallel network model is established.The original leaf image and Gabor texture feature image are input respectively.The extracted features are fused before the output layer,and the recognition results are obtained by softmax function.The accuracy rate of identification is 99.39%.A leaf recognition model of alien invasive plants based on improved vggnet was proposed.The second layer convolution of vgg-16 model is replaced by pyramid convolution layer,which is used to extract contour texture features of different details.At the same time,the pyramid feature module is established.The feature map extracted by pyramid convolution is pooled to reduce the dimension of height and width to form a pyramid feature map.The shallow contour texture features are fused with the features of different network layers.The accuracy rate of identification on the self built leaf dataset of alien invasive plants was 99.63%.The experimental results show that compared with the traditional machine learning method,convolution neural network greatly improves the accuracy of classification and recognition of alien invasive plant leaves.At the same time,the two classification methods based on convolution neural network proposed in this paper can reduce the problem of reducing the recognition accuracy caused by the similarity of different blade edges.