Aarthi Chelladurai ¹ Manoj Kumar D.P ² S S Askar ³ Mohamed Abouhawwash ^4*

• ¹ Department of Electronics and Communication Engineering, Sengunthar Engineering College, Tiruchengode, India
• ² Department of Computer Science and Engineering, Kalpataru Institute of Technology, Tiptur, Karnataka, India
• ³ Department of Statistics, College of Science, King Saud University, Riyadh, Saudi Arabia
• ⁴ Michigan State University, East Lansing, United States

Tomatoes are considered as one of the valuable vegetables over the world due to its usage and minimal harvesting period. However, an effective harvesting still remains as a major issue because tomatoes are easily susceptible to weather conditions and other type of attacks. By understanding this, numerous researches are introduced based on deep learning model for an efficient classification of tomato leaf disease. But, the usage of single architecture does not provide better results due to the limited computational ability and classification complexity. So, this research introduced Transductive Long Short-Term Memory (T-LSTM) with attention mechanism. The attention mechanism introduced in T-LSTM has the ability to focus on various part of the image sequence. The transductive learning exploits the specific characteristics of the training instances to make accurate predictions. This can involve leveraging the relationships and patterns observed within the dataset. The T-LSTM works based on transductive learning approach and the scaled dot product attention evaluates the weights of each step based on hidden state and image patches which helps in effective classification. The data gathered from Plant village dataset and accomplished the pre-processing based on image resizing, color enhancement and data augmentation. Then, these outputs are processed into segmentation stage where U-Net architecture is applied. After segmentation, VGG-16 architecture is used for feature extraction and the classification is done through proposed T-LSTM with attention mechanism. The experimental outcome shows that the proposed classifier achieved an accuracy of 99.98% that is comparably better over existing Convolutional Neural Network (CNN) with transfer learning and IBSA.