98% Accuracy Achieved in autumn Army Worm Detection Using Thermal Fusion

98% accuracy. This is the new standard established by a novel deep learning framework integrating RGB and thermal imaging. For maize farmers, the autumn Army Worm (FAW) is an existential threat, capable of decimating yields rapidly. Current **autumn Army Worm detection** relies on human scouts walking fields. This process is labour-intensive, slow, and frequently inaccurate. By the time a human spots the damage, the crop is often compromised. This study presents an automated solution that removes human error from the equation.

Techniques for advanced autumn Army Worm detection

The researchers constructed a hybrid DNN-ViT model. It addresses the limitations of single-source data. Standard cameras (RGB) capture surface details. Thermal cameras capture physiological stress. Separately, they are useful; together, they are formidable. The framework employs two distinct pipelines. The first utilises feature-level fusion, where Convolutional Neural Networks (CNN) extract specific attributes from both image types before classification via a Deep Neural Network. The second pipeline uses image-level fusion. Here, a composite six-channel image is fed directly into a modified Vision Transformer (ViT). This complex architecture allows the system to 'see' the infestation not just as a visual blemish, but as a thermal anomaly.

Statistical dominance and future application

The results measured in this study are unambiguous. When the model operated without fusion, accuracy collapsed to 0.60. An AUC-ROC of 0.67 indicates a system barely better than chance. However, the fused model achieved 0.98 across all key metrics: accuracy, precision, recall, and F1-score. This drastic differential suggests that multimodal data is not optional for high-reliability monitoring; it is a requirement. The study measured these outcomes on a test dataset, confirming that the fusion of thermal and RGB data significantly outperforms either modality used in isolation. While the lab results are definitive, they suggest that field deployment could face challenges regarding environmental robustness. Future work must validate if this 0.98 accuracy holds up against the chaotic variables of an open maize field.