DFA-SCA NET: Dynamic Feature Amplification with Spatial and Channel Attention Convolution Network for Paddy Disease Detection

A vital staple crop that greatly boosts economies and food security worldwide is paddy. However, there are significant risks to crop quality and productivity from paddy diseases such bacterial blight, brown spots, and leaf muds. To ensure good crop production and efficient management, early and accurate detection of these diseases is crucial. In paddy disease detection, Dynamic Feature Amplification (DFA) offers a major breakthrough by combining global and geographical contexts to improve feature representations in CNNs. Convolutional layers are used in DFA multifaceted approach to extract specific spatial features, Global Average Pooling (GAP) is used to capture global context, and dense layers are used for channel-wise recalibration. Channel-wise recalibration in DFA is accomplished in two stages: first, dense layers with non-linear activation functions are used to create attention weights, which modify the relative value of information on various channels. Secondly, spatial attention processes that highlight important regions in the feature maps, the network is able to recalibrate its spatial context and concentrate on pertinent spatial features. Diseases like Leaf smut, brown spots, bacterial blight, and healthy paddy have all been identified and categorised using the DFA methodology. The model's ability to suppress less significant signals and amplify relevant ones is made possible by the integrated recalibration processes, which improve disease detection accuracy and robustness. In identifying between sick and healthy paddy, DFA distinguished with an amazing accuracy of 91.2%, confirming its efficacy in enhancing model performance.