Poultry Disease Identification Based on Light Weight Deep Neural Networks

Abstract

Poultry farmers are often plagued by poultry diseases during production and face the risk of large-scale spread of poultry epidemic diseases. Accurate and efficient identification of poultry diseases is a necessary prerequisite for timely symptomatic treatment and economic loss avoidance. In this paper, a poultry disease identification model based on a light weight deep neural network is established and named PoultryNet, which adopts MobileNetV3 as the backbone. A feature fusion structure is designed to enhance the feature extraction ability of the model, and a SA module is used to add channel attention and spatial attention. The experiment result shows that the classification accuracy of the proposed PoultryNet for poultry feces images is 97.77%, which is higher than that of MobileNetV3, ShuffleNetV2, EfficientNet, and GoogleNet models by 1.12%, 1.67%, 1.27%, and 3.97%, respectively. Compared with the base model, the amount of parameters of PoultryNet was reduced by 0.33 M. The effectiveness of PoultryNet, as a poultry disease identification model, is therefore proved.