Image Recognition Of Pleurotus Ostreatus Based On Convolutional Netural Network

Pleurotus ostreatus is one of the six major edible mushrooms in China,and its annual consumption and export volume are very considerable.Under the background of the factory production of edible fungi,the automatic cultivation and management of Pleurotus ostreatus have been well realized.However,due to the irregular shape of the Pleurotus ostreatus,the delicate and fragile physical properties of the fruiting body,whose automatic harvesting technology is always difficult to achieve.In order to prevent the automatic harvest equipment to destroy the fruiting body,it is necessary to study the accurate identification of Pleurotus ostreatus.First,the collected images are filtered and sorted out.After removing the unqualified images,the areas of relative concentration of the target distribution in the remaining images are intercepted and zoomed,making the size of them equal to 400*400.Take 600 processed images as the positive sample data set,and then intercept 400 representative regions with the same background processed by the same method,the images processed by unified zooming are taken as negative samples with the positive samples used together as neural network to identify the data set.Eighty percent samples are randomly selected as training set,and the other twenty percent are test set.In addition,picking up the characteristic parameters as test preparations is by means of gray scale,background segmentation and image filtering.The main work of this paper is to identify the collected Pleurotus ostreatus images by the convolutional neural network.Through the improvement of the classic LeNet-5 network model,an improved model suitable for the identification of Pleurotus ostreatus is obtained.The number of convolutions,the number and size of convolution cores,and the size of sampling units in the model are adjusted and modified.Taking the recognition correct rate as the only index,this paper arranges test for these three sets of network parameters,learning rate,learning batch and training times to get the optimum parameter combination.A convolutional neural network model is built on MATLAB platform.480 positive samples and 320 negative samples are used as training set,120 positive samples and 80 negative samples as test set.It gets the recognition correct rate for the data set of the improved model as 97.25% and 94.5% and the average time for recognition as 1.078 s and 1.117 s.At the end of this paper,BP neural network with a better ideal recognition effect is selected to identify Pleurotus ostreatus.It gets the recognition correct rate of training set and test set as 93.38% and 91.5% and the average time for recognition as 1.526 s and 1.586 s.By comparing the recognition results with convolutional neural network,the recognition correct rate of convolutional neural network for two data sets is 3.87 and 3 percentage higher than that of BP neural network,and the average time saving is 0.448 s and 0.469 s respectively.It is known that the performance of the convolution neural network is better than the traditional BP neural network,which is more suitable for the identification of the Pleurotus ostreatus image,whether it is in the correct rate or the speed of recognition.