Computer Vision and Pattern Recognition, Artificial Intelligence, Computer Science Applications, Agricultural and Biological Sciences
12
Scopus Publications
Scopus Publications
A Hybrid Adaptive Dual Attention Network for Precise Classification of Nutrient Deficiencies in Agricultural Crops International Journal of Intelligent Engineering and Systems, 2026 Nutrient deficiencies in crops are a significant challenge in agriculture.Lack of essential nutrients directly affects both crop yield and quality.To address this problem, a hybrid model, Adaptive Dual Attention Network (ADAN), is proposed.ADAN combines the strengths of Densenet121 and MobileNet using dynamic routing feature fusion, which adaptively blends features from pretrained networks to overcome the limitations of conventional feature fusion methods.The fused features are further refined using the attention mechanism-channel attention and multiscale spatial attention-to enhance feature representation and better handle classes with similar visual symptoms.To improve the visibility of localized symptoms in the input images, domain-specific pre-processing using CLAHE is applied.ADAN and conventional feature fusion models are evaluated on the Mulberry and the Rice datasets using 5fold cross-validation.The reported results are the aggregate performance across all five folds.On the Mulberry dataset, ADAN achieves 94.21% accuracy and 93.85% F1 score.ADAN demonstrates better performance in classifying visual overlap classes, Nitrogen, Sulfur, and Iron.On the Rice dataset, the accuracy is 98.01%, and the F1 score is 97.94%.Grad-CAM visualizations validate ADAN's attention-based interpretability.
Land Use Classification using Ensemble Hybrid Model: A Study on the UC Merced Dataset Shrihari V Pandurangi, Varshini L, Yashas N, Tejas M Bharadwaj, S Nikitha, Prabhanjan Soukar 2025 International Conference on Computing Technologies Icoct 2025, 2025 Accurate land-use classification from remote sensing imagery plays a crucial role in environmental monitoring, urban planning, and disaster management. This study presents an efficient deep learning-based approach for classifying land-use patterns using high-resolution aerial images from the UCMerced Land Use dataset. We evaluate multiple pretrained convolutional neural networks (CNNs), including MobileNet, DenseNet121, VGG16, and VGG19, comparing their performance in extracting discriminative spatial features.To enhance classification accuracy, we propose a hybrid feature fusion model that combines the strengths of MobileNet’s lightweight architecture and DenseNet121’s dense feature reuse. The extracted deep features are processed using a linear-kernel SVM, while SMOTE oversampling ensures balanced class representation. Our experiments employ stratified 5-fold cross-validation to validate model robustness. The results demonstrate that the hybrid fusion mode achieves the highest accuracy (96.57%), followed by the MobileNet-based model (95.48%). Notably, the proposed approach maintains computational efficiency, making it suitable for real-time applications. Detailed confusion matrix analysis reveals common misclassifications, providing insights for future improvements. This work contributes to advancing automated land-use mapping by optimizing deep feature extraction and classification techniques for remote sensing applications.
COMPOSITE CROSS ATTENTION NETWORK FOR RELIABLE AND ROBUST PLANT NUTRITIONAL DEFICIENCY ANALYSIS: IMAGE ENHANCEMENT AND TRANSFER LEARNING Journal of Theoretical and Applied Information Technology, 2024
Handwritten electric circuit diagram recognition: An approach based on finite state machine Lakshman Naika R, , Dinesh R, Prabhanjan S International Journal of Machine Learning and Computing, 2019 In this paper we propose a method for recognizing hand drawn electronic circuit diagrams. The proposed method first detect and classify each components present in the hand drawn circuit diagram. For the purpose of component recognition, we have constructed the feature vector by combining Local Binary Pattern (LBP) and statistical features based on pixel density. Classification of components is done by using support vector machine (SVM) classifier. Upon detection and recognition of components, the proposed method subsequently uses the position and sequence of arrangement of components to determine the type of circuit. For the purpose of establishing the sequence of components we have used finite state machine. The proposed method represents the sequence of recognized components as a string. This string representation of circuit is fed to a Finite State Machine (FSM) to detect type of circuit. The proposed method has been tested on about 100 hand written circuit diagrams of varying complexities and of different types. The proposed component detection method gives over 99% accuracy whereas, the circuit recognition method has recognition rate of over 85% recognition rate for the circuit type recognition.
Deep Learning Approach for Devanagari Script Recognition S. Prabhanjan, R. Dinesh International Journal of Image and Graphics, 2017 In this paper, we have proposed a new technique for recognition of handwritten Devanagari Script using deep learning architecture. In any OCR or classification system extracting discriminating feature is most important and crucial step for its success. Accuracy of such system often depends on the good feature representation. Deciding upon the appropriate features for classification system is highly subjective and requires lot of experience to decide proper set of features for a given classification system. For handwritten Devanagari characters it is very difficult to decide on optimal set of good feature to get good recognition rate. These methods use raw pixel values as features. Deep Learning architectures learn hierarchies of features. In this work, first image is preprocessed to remove noise, converted to binary image, resized to fixed size of 30[Formula: see text][Formula: see text][Formula: see text]40 and then convert to gray scale image using mask operation, it blurs the edges of the images. Then we learn features using an unsupervised stacked Restricted Boltzmann Machines (RBM) and use it with the deep belief network for recognition. Finally network weight parameters are fine tuned by supervised back propagation learning to improve the overall recognition performance. The proposed method has been tested on large set of handwritten numerical, character, vowel modifiers and compound characters and experimental results reveals that unsupervised method yields very good accuracy of (83.44%) and upon fine tuning of network parameters with supervised learning yields accuracy of (91.81%) which is the best results reported so far for handwritten Devanagari characters.
Handwritten Devanagari characters and numeral recognition using multi-region uniform local binary pattern Prabhanjan S, R Dinesh International Journal of Multimedia and Ubiquitous Engineering, 2016 Automated offline handwritten character recognition of Devanagari script is a growing area of research in the field of pattern recognition. A new approach for Devanagari handwritten character / digit recognition has been proposed in this paper. This approach employs Uniform Local Binary Pattern (ULBP) operator as the feature extraction method. This operator has great performance in research areas such as texture classification and object recognition, but it has not been used in Devanagari handwritten character/digit recognition problem. The proposed method extracts both local and global features. The proposed method have two steps, in the first step image is preprocessed to remove noise and to convert it to binary image and then resizing it to a fixed size of 48x48. In the second step, ULBP operator is applied to the image to extract global features then input image is divided into 9 blocks, ULBP operator is applied to each block to extract local features. Finally, global and local features are used to train Support Vector Machine(SVM). The proposed method has been tested on large set of handwritten character and numeral database and empirical results reveals that the proposed method yields very good accuracy (98.77%) . To establish the superiority of the proposed method, it has also been compared with the contemporary algorithms. The comparative analysis shows that the proposed method out performs the existing methods.
A novel method for document skew detection and correction: Application to, handwritten document and bank documents International Journal of Applied Engineering Research, 2015
Classification of structured documents: An approach based on surf features International Journal of Applied Engineering Research, 2015
