Research On Multi-feature Fusion Neural Network In Plant Leaf Recognition

Plants are widely distributed throughout the world and play an important role in ecosystem energy flow and material circulation.At the same time,understanding plants and excavating their own values is of great significance to promoting national economic development and improving agricultural productivity.Plant leaves as the main organ of plants are also the main reference objects for distinguishing plant species.Early leaf recognition methods stayed in the morphological comparison stage.With the continuous development of image processing technology,researchers shifted their attention to leaf image feature extraction and analysis.Extracting reasonable feature quantities on plant leaf images and designing matching classifiers has been a hot topic in this field.It is in this context that this paper proposes two high-dimensional leaf features quantity extraction algorithms,Fourier-curvature space descriptor and CLDgray gradual change matrix,and applies it to multi-feature fusion neural network to realize leaf recognition.This paper studies involving follows:1.Starting from image shape,color and texture,the existing image feature extraction algorithm is studied and applied to leaf image feature extraction.The recognition effect is tested and analyzed under the existing leaf library.2.A high-dimensional combined feature quantity extraction algorithm for leaf images is proposed.Under the same leaf library and recognition algorithm,the combined feature recognition effect is better than the existing image feature extraction algorithm.3.Construct a neural network classifier suitable for the feature quantity of the leaf image in this paper to further improve the recognition effect.4.Comparative recognition algorithm include CNN under time complexity(time extraction of feature extraction calculation),space complexity(mem-ory usage),network training time,global average recognition rate,average singleton identification time-consuming,total number of calculation param-eters,etc.The main purpose of this study is to design a smaller leaf recognition method than the convolutional neural network.Under the premise of ensuring the recognition effect,the calculation amount can be minimized,the network depth can be reduced,and the recognition task can still be completed on the platform with lower computing power.From the relevant experimental data in this paper,although the recognition effect is weakened under the larger-scale leaf bank,the relevant evaluation indicators can reflect the superiority of the method.In addition,the content of this paper is intended to point out that in some image recognition tasks,it is not necessary to rely entirely on neural networks,and combining traditional image processing with neural networks can often achieve better results.