The Identification Of Crop Diseases Base On Deep Learning

With the rise of precision agriculture,the accurate identification of diseases is also attracting much attention.Faced with the high requirements of real-time and accuracy for accurate identification,the traditional digital image processing technologies have their limitations,the manual design features are required,and mean power and material resources are consumed.And it just can identify single crops and it doesn’t work well.Therefore,it is very meaningful to find an effective and real-time disease image recognition method.Deep learning is a machine learning algorithm which can learn representative features automatically and make it possible to achieve good results in the field of image recognition.This paper researches crop disease image recognition technology based on deep learning and the mainly jobs are as follows:The object of this study are wheat powdery mildew and leaf rust,peanut black spot and brown spot disease,tobacco wildfire disease,brown spot disease and mosaic disease collected in the field environment.A total of 5,900 images were collected and the disease were identified by using a deep convolutional neural network in a complex background.On the basis of the convolutional neural network,three deep convolutional neural network models(Alex Net,VGG-16,and Goog Le Net)were used and combined with migration learning and fine-tuning for the image recognition of the seven diseases and the three types of crops.The recognition results of different algorithms and different iteration times were used.The results of average recognition accuracy,various disease recognition rates,recall rates,recall rates,and F1 values showed that the use of VGG-16 model was effective and the average disease recognition rate was reached98.92%.It shows that it is feasible to use deep learning to identify diseases in crops,identify a variety of crop diseases,and solve the problems of artificial extraction characteristics;At the time of training,150 rounds have converged,indicating that the use of migration learning and fine-tuning training methods can reduce learning time and reduce training complexity on small datasets.