Panoramic Plant Recognition Method Based On CNN And GLCM Fusion Discrimination

Plant classification is one of the basic research contents in system biology.With the development of artificial intelligence technology,there are many kinds of plant recognition algorithms.Because the leaves of each plant are different and contain many texture features.Therefore,most of the common plant identification methods are to identify the leaves,but this recognition can only be used for academic research,and has little significance in practical application.In the model training of plant image classification directly,the recognition rate of test set is poor or over fitting phenomenon will appear due to the appearance similarity of plants and the color and shape changes caused by environmental impact.In order to solve the above problems,this paper proposes a plant recognition method based on DS evidence theory fusion of convolutional neural network(CNN)and gray level co-occurrence matrix(GLCM).Firstly,a variety of preprocessing operations are carried out on the plant image,including grayscale processing,clipping compression and size normalization,so as to reduce the amount of data and unify the sample specifications of the dataset.In order to highlight the texture features of plant leaves,the local two-dimensional model(LBP)is used to extract the texture information from the image.In addition,due to the sparsity and high-dimensional characteristics of LBP feature data,principal component analysis(PCA)method is used to further extract the effective content of the data,and reduce the data dimension to obtain more efficient LBP features.Then the LBP feature and gray image classification model are trained based on convolution neural network and gray level co-occurrence matrix.Finally,DS evidence theory is used to combine the recognition results of the two models to obtain better classification effect.In the first mock exam,the two problems of poor recognition rate of single model and low practical application value of traditional plant leaf identification are as follows:(1)This paper introduces the block complete LBP feature,namely CLBP theory.CLBP theory improves the original LBP Operator,which not only has the characteristics of high dimension and sparsity,but also can use the method of dimensionality reduction to compress data and facilitate the subsequent calculation.In addition to LBP Operator,the absolute amplitude value of each pixel and the gray value of central pixel can be extracted.These features also provide effective identification information.(2)GLCM has many features.This paper introduces that GLCM has five typical features,and tests show that the selection of energy,contrast,inverse moment and entropy can make the recognition efficiency higher.Based on this,the original gray level co-occurrence matrix algorithm is improved.(3)This paper mainly focuses on the recognition of plant panoramic images,which can better eliminate the interference of light and dust and other complex background.Compared with the identification of plant leaves,it has higher recognition difficulty and has greater practical application value.In addition,in the discussion of experimental results,the recognition scheme described in this paper is compared with several representative classification methods,such as support vector machine(SVM),k-means(K-means),BP neural network,etc.By comparing the accuracy,purity,F1 value and adjusted Rand coefficient(ARI)of these methods,it is verified that the scheme has high generalization ability and recognition effect.Finally,the process and characteristics of this scheme are summarized,and the further direction and application prospect are prospected.