CNNAttLSTM: An Attention-Enhanced CNN– LSTM Architecture for High-Precision Jackfruit Leaf Disease Classification

Gaurav Tuteja¹Fuad Ali Mohammed Al-Yarimi²Amna Ikram³Rupesh Gupta¹Ateeq Ur Rehman^4*Jeewan Singh^4,5Irene Delgado Noya⁶LUIS ALONSO DZUL LOPEZ⁶

• ¹Chitkara University, Rajpura, India
• ²King Khalid University, Abha, Saudi Arabia
• ³The Government Sadiq College Women University Bahawalpur, Bahawalpur, Pakistan
• ⁴Gachon University, Seongnam, Republic of Korea
• ⁵Chandigarh University University Centre for Research & Development, Sahibzada Ajit Singh Nagar, India
• ⁶Universidad Europea del Atlantico, Santander, Spain

Abstract: Jackfruit cultivation is highly susceptible to leaf diseases that reduce yield, quality, and farmer income, yet early diagnosis remains difficult due to the limitations of manual inspection. This study presents a hybrid deep learning framework, CNNAttLSTM, which integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) units, and an attention mechanism for multi-class classification of algal leaf spot, black spot, and healthy jackfruit leaves. The proposed methodology begins with data preprocessing, where all jackfruit leaf images are resized to 224×224 pixels, normalized to the [0,1] range, and divided into ordered 56×56 spatial patches treated as pseudo-temporal sequences. These patches are processed through sequential Conv2D, MaxPooling2D, and GlobalAveragePooling2D layers for spatial feature extraction, followed by LSTM units to capture sequential dependencies, and an attention mechanism that assigns adaptive weights to emphasize diagnostically significant regions before final classification. Although the dataset consists of static images, each image is divided into ordered spatial patches that are sequentially processed by the LSTM to learn contextual dependencies across different regions of the same leaf. This approach enables temporal modelling of spatial feature relationships, effectively treating patch sequences as pseudo-temporal data. Experiments were conducted using the publicly available Kaggle jackfruit leaf disease dataset containing 38,019 images, maintaining its predefined training, validation, and testing splits. The proposed CNNAttLSTM achieved 99% classification accuracy, outperforming baseline CNN (86%) and CNN-LSTM (98%) models while remaining computationally efficient. The architecture contains only 3.7 million parameters—an 18% reduction compared to CNN-LSTM—converges within 45 minutes of training on an NVIDIA Tesla T4 GPU, and performs inference in 22 milliseconds per image (batch size = 1). These metrics demonstrate low computational cost and confirm the model's suitability for real-time, mobile, and edge-device deployment. Overall, the findings highlight CNNAttLSTM as a robust and efficient solution for automated jackfruit leaf disease detection and broader precision agriculture applications.