| With the rapid development of digital image processing technology and computer vision,plant image identification has become a focus in trans-disciplinary research in the botanical taxonomic.Most image-based identification methods proposed in the past were so far based on leaf or flower,lacking of overall characteristics of plants,and identifying the images of single background.With the popularity of smartphones,explosive growth of the conventional plant images,the few characteristics extracted by traditional handcraft feature engineering and the classifier with simple structure lead to the low recognition rate,when plant species increase to a large number.To overcome aforementioned challenges,the research based on residual network with deep learing algorithm is proposed,that is significant for classifyinig large-scale datasets.Due to the lack of validated conventional plant image dataset collected by mobile phone,the conventional plant image dataset BJFU100 collected by mobile phone in natural scene is presented,which contains 10,000 images of 100 ornamental plant species in Beijing ForestryUniversity campus.A 26-layer deep learningmodel consisting of 8 residual building blocks is designed,that contains enough trainable parameter to learn the discriminative feathers.ResNet26 results in fast and robust convergence during SGD optimization,which prevents overfitting or falls into local optimum.Comparing with the original ResNet model,ResNet26 has a very good recognition rate in the BJFU100 set,resulting in 91.78%accuracy.To show the effectiveness of the proposed ResNet26 model,a series of experiments have been performed on the publicly available Flavia leaf dataset of single background.ResNet26 model achieves a 0.28%improvement compared with the best-performing method.In addition,this study also develops Web applications and WeChat Mini applications based on deep learning,implementing plant recognition of website and mobile phone,and promoting the further development of smart forestry. |
PDF Full Text Request