Accurate and efficient plant disease detection is crucial for precision agriculture to help minimize yield losses. Nevertheless, existing deep learning models are computationally resource-intensive, making them unsuitable for real-time monitoring with drones. In this work, we propose a lightweight vision transformer with a token pruning and knowledge distillation approach to detect plant diseases. We follow a teacher-student architecture where a vision transformer with complete feature learning ability serves as a teacher model and guides a student vision transformer. Attention-based token pruning helps prune out unimportant patch tokens from the student network during transformer layers and decreases computational redundancy without removing disease-relevant tokens. The proposed approach is evaluated on the PlantVillage color dataset with 54,305 images belonging to 38 classes (disease and healthy). Our experimental study shows that the proposed Student Vision Transformer network achieves a remarkable accuracy score of 96.47% and a macro F1-score of 95.30%, using only 1.28 million parameters. Moreover, our approach decreases the number of parameters by 61%. Our robustness tests under various transformations demonstrate the applicability of our vision transformer with pruning and knowledge distillation for real-time drone-assisted crop disease monitoring.
D. M, Sweetlin Jebakani D, J. J· 2026 4th International Confe...· 0 citations
The rise of digital payments has magnified credit card fraud issues the complexity, scope and therefore the attack strategies have evolved to become an enormous obstacle for the traditional, static (rule based) and single- modal methods of ML in learning dynamic transactional patterns. In this paper we introduced a multimodal fraud detection approach combining the typical numerical features characterizing single transactional entities with their time- series transaction behaviour so that we could achieve higher accuracy in identifying frauds. The former represents properties for each transaction; temporal patterns in user spending are captured by a Gated Recurrent Unit (GRU), a network effectively modelling sequences of events (purchase history in our context) and then combined to finally perform binary classification in which either fraudulent or real transaction class is detected. We use the public domain 284,807 transaction records from credit card dataset; fraud transactions account for 0.172 of the samples. The experiment shows our novel framework (multimodal GRU) outperformed existing ML algorithms and single-mode techniques, having the recall reach 0.89, F1-score value 0.90 and AUC score 0.99. It’s highly scalable and efficient to use in real time monitoring applications. Our proposed method of combining two types of features: a set of single number characteristics and sequence-based information to successfully solve this challenging problem in modern finance.
J. J, S. S, Sabari Rr· 2026 International Conferenc...· 0 citations
