Jujube varieties recognition based on multi-scale convolution and attention mechanism

In order to improve the accuracy of jujube varieties recognition method in natural scenes, a jujube varieties recognition model（Jujube-ResNet-18） was proposed by integrating multi-scale convolution and attention mechanism. In this study, ten types of jujube varieties under natural scenes were taken as objects. According to the characteristics of jujube variety images, the model in this paper was improved on the basis of ResNet-18. Firstly, the multi-scale convolution module was introduced to enhance the ability of the model to extract multi-scale features of jujube fruit. Secondly, the attentional mechanism CBAM was added into each residual block to increase the weight of jujube fruit feature information and weaken the influence of complex background and other useless features. The experimental results showed that the accuracy of Jujube-ResNet-18 on the date variety dataset was 89.5%, while the number of parameters and weight were only 1.135×10~7 and 43.41 MB, respectively. Compared with other algorithms, Jujube-ResNet-18 has better feature extraction ability, anti-interference ability and smaller model complexity, which can provide a reference for the study of jujube varieties recognition in natural scenes.