Design And Analysis Of Crop Leaf Recognition System Based On Deep Learning

Since ancient times,our country has been a big country that focuses on agricultural production.Therefore,the identification and monitoring of crops,as well as further scientific management methods,have an extraordinary significance for the good growth of crops and the increase of yield.In this case,it is necessary to implement a stable system that can automatically identify the classified crops with a computer.At the same time,the gradual popularization and vigorous development of mobile intelligent terminals has gradually increased the requirements for corresponding identification technologies.Therefore,the accurate identification of crops also has certain commercial value.Traditional blade recognition methods include main processes such as preprocessing,feature extraction,and feature matching.Due to the diversity of the blade shape,similarity,illumination differences,background factors and other issues,the effect of classification will be seriously affected.Because it is artificially extracting the features of the blade and marking the blade,it takes a lot of time and effort.At the same time,errors are also generated during manual operation,which affects the final accuracy.In recent years,the rapid development of artificial neural networks has played an important role in plant leaf recognition.The advantages of artificial neural network are: high accuracy classification,strong learning ability,and good fault tolerance for image noise.For example,BP neural networks are more accurate in plant leaf recognition than traditional leaf recognition methods.By pre-processing the image,and then reducing the dimension of the input data,a large number of calculations are reduced.Finally,the fully connected layer performs autonomous training on the reduced-dimensional data to form a classification.However,when the BP network processes the data autonomously,the data is processed too many times,which will make the convergence speed of the whole network slower or even over-fitting,and the accuracy of plant leaf recognition is reduced.In recent years,deep learning has become more and more important in the direction of image recognition.Convolutional neural networks provide a better solution for plant leaf recognition technology.This paper proposes a crop leaf recognition method based on convolutional neural network.Compared with the traditional method,the feature extraction and classifier are simplified,which reduces the influence of external factors on feature extraction.When the data set was created,nearly 8,000 photos of 24 types of crops were directly photographed,and after data enhancement and de-weighting,they became 15,000 or so,and the crop leaves could be displayed more comprehensively.Five common convolutional neural network models were then selected for appropriate improvements to reduce over-fitting and reduce the number of parameters in the model.Good accuracy was achieved in the separate tests.Finally,based on the model fusion technology,the five improved models are merged into one model.In the same data,the accuracy is higher than that of a single model.In the contrast experiment,I chose Densenet neural network.Under the same experimental environment and data,the accuracy of the proposed model is higher than that of Densenet neural network,which is 96.12%.