AppleSENet: Apple Leaf Disease Classification Using Lite SE-Network

Leaves have an important place in the early diagnosis of diseases in plants. In the fight against leaf diseases and pests, technology increases efficiency by producing faster and more accurate results. In addition, chemical pesticides are used at more accurate rates through technology-based determinations, contributing to the ecological balance. In this study, the AppleSENet deep network model is proposed for the detection of apple leaf diseases. In this model, channel attention weights were produced by integrating the SE block between the convolution layers. These weights express the importance of the information carried by the channels. In this way, it is aimed to improve performance by highlighting the attributes that are effective in solving the problem. The proposed AppleSENet model showed higher classification performance ($\mathbf{9 9 . 2 1 \%}$) with fewer parameters compared to SqueezeNet, ShuffleNet and MobileNetV2 deep networks, which have shown successful results in the literature.