Classification of Birds Based on Weighted Fusion Model
Peng Wu, Xiaomei Yi, Lufeng Mo, Gang Wu · 2021
Fine-grained image recognition is one of the research highlights of deep learning, and bird recognition is a classic application of fine-grained image recognition. Because different kinds of birds are very similar in appearance, so it is a common method to verify the fine-grained recognition model through the bird recognition effect. This paper chose CUB-200-2011 bird data set and VGG16 and ResNet152 were selected as the basic models. And by calculating the fusion between different weights and different models to obtain optimal classification weighted fusion model for data sets. Model weights can be adjusted dynamically, which makes transfer learning more flexible in weight allocation and stronger model generalization. For the constructed model, Adam optimizer and SGD optimizer are respectively used to analyze and compare the four different learning rates of 0.1, 0.01, 0.001 and 0.0001, to obtain the weight parameters with the best classification effect.