Toadstool Image Recognition Based On Deep Residual Network And Transfer Learning

Various toadstools spreading widely in China are often taken by mistake because people lack knowledge of distinguishing them from edible fungi.Existing methods of toadstool identification exist or rely on personal experience is not completely accurate,or need very professional knowledge of mycology and laboratory equipment defects,cannot be popularized.In order to reduce the occurrence of people’s health or even life safety threatened by accidental ingestion of toadstool,an accurate,effective,convenient and quick identification method against common toadstool is needed.Based on the above research background and current situation,by establishing the data set of common toadstool images in China,this paper proposes to use deep residual network and transfer learning technology to classify and identify toadstool images.The research work is as follows:(1)In view of the current lack of high-quality available public data set of toadstool images,according to the common species of toadstool listed in the Computer Network Information Center of Chinese Academy of Sciences,a total of 11695 toadstool image data sets of 18 species were established through collection and collating toadstool images.(2)Proposed based on the depth of the residual network and transfer learning toadstools id entification method,this method takes the Res Net-152 network as the pre-training model,the model-based transfer learning method and Adam algorithm as the model optimization method w ere applied to construct the model structure of toadstool image recognition.Finally,the model t raining was carried out by the k-fold cross validation.And thirteen different comparative experi ments were designed to explore the influencing factors of toadstool image recognition results.(3)Based on the toadstool recognition model proposed in this paper,a toadstool image recognition system on Android platform is designed and implemented,which has the functions of toadstool image collection,image uploading,identification and display of identification results.The experimental results show that the recognition rate of common toadstool images is high.The accuracy of Top-1 and Top-5 of toadstool image recognition model even can reach 92.17% and 97.35% respectively,which reaches the research goal of this paper and has certain research significance.The toadstool image recognition system designed and implemented in this paper has good accuracy and timeliness in the identification results of toadstool images uploaded by users,and is simple to operate,widely applicable,with certain practical significance and social value.