Bird Recognition Based on Mixed Convolutional Neural Network

Bird image recognition is a classic experiment in the fields of artificial intelligence and machine vision. In order to better explore the recognition patterns and performance differences of mainstream convolutional structures nowadays, an experiment based on mixed convolutional neural networks was designed. In the experimental model design section, the hybrid convolutional neural network focuses on dense connections and adopts residual connections in dense connections to ensure that the network is easier to learn. In order to ensure that the network extracts more feature information, an Inception structure based on deep separable convolution is also introduced. Afterwards, an improved channel attention mechanism is used to learn the weights between channels and an adaptive convolution structure is used to learn spatial weights. In comparative experiments conducted on a 525 classified bird dataset, the accuracy of the mixed convolutional neural network experimental model was 94% with fewer parameters, which was superior to models such as Inception V2, ResNet101, DenseNet264, MobileNet V3-large, and EfficientNet. This indicates that it has certain performance advantages in bird feature recognition tasks.