Optical Memory and Neural Networks最新文献

{"title":"Design and Implementation of Coconut Eriophyid Plant Disease Detection Using Mask-RCNN","authors":"Muhammed Ilyas H.,&nbsp;M. Sudheep Elayidom,&nbsp;Daleesha M. Viswanathan","doi":"10.3103/S1060992X25600880","DOIUrl":"10.3103/S1060992X25600880","url":null,"abstract":"<p>One of the most significant perennial tropical crops is the coconut palm (<i>Cocos nucifera</i> L). The coconut plantation sector relies on specialist guidance to diagnose and control diseases. Plant disease detection also uses spectroscopic techniques such as Ultraviolet-Visible (UV) spectroscopy, mass spectrometry, and nuclear resonance spectroscopy. Deep learning (DL) technology in computer vision has opened up a new path for finding agriculture solutions. Plantations of coconuts are seriously threatened by coconut eriophyid infections, which can result in severe output reductions. For prompt intervention and efficient disease treatment, accurate disease identification is essential. In this article, we offer a unique method for identifying the coconut Eriophyid plant disease, which combines Variational Autoencoders (VAEs) for noise reduction with Mask Region-based Convolutional Neural Network (Mask R-CNN) for accurate disease segmentation. The developed trained model’s performance was examined.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"574 - 587"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098351","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel Siamese CNN-Transformer Architecture for Efficient and Robust Cross-Domain Image Recognition","authors":"Ali Zeydi Abdian,&nbsp;Mohammad Masoud Javidi,&nbsp;Najme Mansouri","doi":"10.3103/S1060992X25600429","DOIUrl":"10.3103/S1060992X25600429","url":null,"abstract":"<p>Machine learning models often struggle with limited training data, particularly in multi-domain scenarios where stylistic variations and data scarcity pose challenges. Although Siamese neural networks have shown promise in resource-limited settings, their generalization ability across various image processing domains, especially for culturally specific scripts (e.g., Persian/Arabic numerals), remains understudied. This study introduces a Siamese-CNN with a multi-scale dynamic self-attention module, designed to enhance feature extraction and cross-domain generalization across various handwritten digit (Persian/Arabic, English), and fashion item datasets. The proposed framework integrates five distinct Siamese-CNN architectures, incorporating multi-scale attention mechanisms and dynamic weight-sharing strategies to improve feature discrimination in low-data regimes. The attention module refines feature representations adaptively, mitigating overfitting while capturing fine-grained details across different styles and scripts. Extensive experiments on the Hoda dataset, a benchmark with complex stroke structures and cultural specificity, demonstrate the superiority of the proposed model, achieving 98.34% accuracy a 6.01% improvement over previous Siamese approaches while maintaining high precision (99.32%), recall (97.35%), and F1-score (98.32%). With an average epoch time of 16.26 s, the model is computationally efficient, making it suitable for real-time applications. This study advances Siamese-CNNs for cross-domain image recognition by: (1) introducing a multi-scale adaptive attention strategy for feature refinement, (2) validating generalization across eight datasets, and (3) demonstrating the role of dynamic weight-sharing in improving representation learning under constrained resources. The proposed framework outperforms conventional methods in both accuracy and efficiency, paving the way for enhanced image recognition in handwritten text digitization, multilingual document analysis, and real-time visual perception tasks.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"481 - 507"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098384","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Application of Color Matching in Chinese Painting Creation Based on a Big Data Analysis Framework","authors":"XiaodongJi","doi":"10.3103/S1060992X26601727","DOIUrl":"10.3103/S1060992X26601727","url":null,"abstract":"<p>The combination of traditional art and artificial intelligence creates new opportunities for cultural heritage innovation and preservation. In order to capture both spatial colour patterns and global contextual links, this study combines convolutional neural networks (CNNs) with multilayer perceptrons (MLPs) to propose a colour matching system for Chinese art production based on a big data analysis framework. A sizable dataset of Chinese paintings was gathered and prepared from public art sources and museum archives. The system provides very accurate colour matching using deep learning, namely the suggested MLP + CNN hybrid model. Metrics like ΔE2000, SSIM, and top-5 accuracy are significantly improved over traditional models. The model serves as a basis for intelligent painting tools that adhere to the aesthetic standards of traditional Chinese art in addition to improving artistic integrity in digital reproduction.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"508 - 519"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098375","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing Image Security with Chaotic Neural Networks and Camellia Encryption Integrated with Steganography","authors":"Abhishek Saini,&nbsp; Ruchi Sehrawat","doi":"10.3103/S1060992X26700098","DOIUrl":"10.3103/S1060992X26700098","url":null,"abstract":"<p>Securing multimedia data, particularly images, has become paramount in the digital era due to the exponential growth of data and increasing cyber threats. This study proposes an innovative hybrid encryption model that integrates chaotic neural networks, specifically convolutional neural networks (CNNs), with the chaotic properties of the Lorenz attractor and the Camellia block cipher, augmented by steganographic techniques. The proposed method generates 16 distinct 256-bit keys from the original image using a CNN, with the first key initiating the encryption process. The Lorenz attractor sequence shuffles image pixels, disrupting their spatial correlations, while a linked-list steganographic approach conceals each subsequent block’s key within the preceding block. Here we show that the encrypted images exhibit high randomness and strong resistance to statistical and differential cryptanalysis attacks, as evidenced by near-ideal entropy values (close to 8), low correlation coefficients (close to zero), and robust differential analysis metrics (NPCR values exceeding 99.57% and UACI values above 33.39%). These results demonstrate the superiority of our approach over existing methods, underscoring its potential for securing image data in various real-world applications.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"634 - 664"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098379","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid AI-Driven Very Large-Scale Integration Floorplan Optimization Using Graph Neural Networks and Reinforcement Learning","authors":"Harshada Hemant Vaidya,&nbsp; Vaidehi Deshmukh","doi":"10.3103/S1060992X26700104","DOIUrl":"10.3103/S1060992X26700104","url":null,"abstract":"<p>Floor planning is an important design phase in Very Large-Scale Integration (VLSI) circuits, as it approximates the space of a chip according to the optimum layout of blocks and their interconnections. This process entails figuring out the locations, shapes, and size of components that make up a chip. Floor planning in VLSI is not an easy task considering that, a slight change in placement of the components can consumes a lot of energy and heat during processing of the signals inside the VLSI architecture. Thus, to propose a framework based on the Hybrid model comprised of graph neural networks and reinforcement learning methods in order to achieve optimized floor planning to reduce the heat and the size of the architecture. Initially, the benchmark GSRC floorset dataset with information about blocks, netlists and connectivity is taken into account. Then, the initial placement of blocks are processed by the Graph Neural Network (GNN) based on identifying the relations among the blocks, netlists and connectivity for all the components to make predictions for the placement of the components in the regions. These initially placed parameters are then optimized using the Deterministic Policy Gradient (DDPG) agent that progressively tries and refines component layouts to improve the floorplan. The optimization process also aims to minimize important parameters such as total area, wire length, power usage and timing delays. The agent advances the layout by refinement in a series of steps until the layout is optimized to such a high level of efficiency and optimization that it satisfies the design goals, and enhances overall performance. Proposed method gives wire length of 15.3080 µm, Area of 519 400.1250 µm² and white space ratio of 0.5732% for n10 circuit respectively. The proposed system learns optimum floorplanning solutions automatically. Its flexibility to various design constraints improves its productivity and extrapolation to VLSI designs.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"454 - 466"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098354","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Neuro-Symbolic Integration of Graph Attention Networks and Knowledge Graphs for Interpretable Personality Prediction from Digital Footprints","authors":"A. Zhdanova","doi":"10.3103/S1060992X26700116","DOIUrl":"10.3103/S1060992X26700116","url":null,"abstract":"<p>This paper addresses the problem of interpretable personality prediction from social media digital footprints. While modern deep learning models demonstrate high accuracy in personality assessment tasks, their black-box nature limits their applicability in psychologically and socially sensitive domains. To overcome this limitation, a neuro-symbolic approach is proposed that combines neural feature extraction from textual and visual data with reasoning over a domain-specific knowledge graph. The architecture integrates Graph Attention Networks to model relationships between observable digital behaviors and personality traits defined by Cattell’s 16 Personality Factors. The study evaluates the proposed approach on social media profiles with psychometric annotations and demonstrates that the model provides competitive predictive performance while enabling transparent and psychologically meaningful explanations.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"440 - 447"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098355","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Early Prediction of ADHD Using LSTM-RNN Model Based on Behavioral Activities","authors":"Deepak Kumar Khandelwal,&nbsp;M. C. Govil","doi":"10.3103/S1060992X24600344","DOIUrl":"10.3103/S1060992X24600344","url":null,"abstract":"<p>Telling apart children who have Attention Deficit Hyperactivity Disorder (ADHD) from those who do not is the central problem this paper takes on, and the tool built for the job layers three pieces together: a recurrent LSTM network that reads behavioral sequences, a Multi-Agent Salp Swarm (MASS) routine that tunes an XGBoost feature-selection stage, and an oversampling method (SMOTE) applied specifically to correct the skew between the two diagnostic groups within the training set. Sequential input is where LSTM networks excel, so that piece is tasked with pulling time-based signal out of the raw behavioral records. What it extracts then passes through MASS-guided XGBoost, a stage whose job is to sharpen how cleanly the two groups separate in feature space. Since ADHD samples are the smaller of the two classes in most such datasets, SMOTE generates additional synthetic minority-class examples before any training happens, so the model isn’t skewed toward predicting the majority class by default. The resulting system achieves an F1-score of 0.99, precision of 0.9980, recall of 0.9952 and overall accuracy of 0.9978 on held-out data. These numbers suggest that the two groups are very cleanly separable under this approach. All in all, these results indicate that the stacking of a sequence-aware deep network on top of a traditional boosted-tree feature selection is a promising approach for early automated ADHD screening, one that could plausibly complement a clinician’s own judgment during diagnosis.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"621 - 633"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel Deep Learning Model for Groundnut Leaf Diseases Classification","authors":"Ritankar Maity,&nbsp;Arunita Das,&nbsp;Krishna Gopal Dhal","doi":"10.3103/S1060992X25603434","DOIUrl":"10.3103/S1060992X25603434","url":null,"abstract":"<p>Groundnut, an essential oilseed crop in India, is extremely vulnerable to a range of diseases that can have a considerable impact on its productivity and quality. Conventional disease detection methods typically involve a significant amount of manual work and necessitate specialized expertise, which limits their practicality for general application in agriculture. This work presents GroNet, which is a lightweight version of Inception-ResNet model for classifying groundnut leaf diseases. The model integrates the inception blocks from InceptionV3 with the skip connections from ResNet to improve feature extraction and address the issue of the disappearing gradient. The performance of the model was assessed using a widely recognized groundnut leaf dataset, comparing it to ResNet152, ResNet50, InceptionV3, and MobileNetV2. The experimental results indicate that GroNet outperforms the other examined models, achieving a superior performance with an accuracy that surpasses 99%. The accuracy of the GroNet highlights its potential as an essential tool for farmers and agricultural experts, facilitating prompt and accurate disease identification and therefore enhancing crop management and production.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"539 - 558"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098381","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Combination of CNN and SVM Classifier for Arabic Handwritten Recognition","authors":"El Mamoun Mamouni,&nbsp;Ismail Bahri,&nbsp;Moussa Laib","doi":"10.3103/S1060992X24600903","DOIUrl":"10.3103/S1060992X24600903","url":null,"abstract":"<p>Handwriting recognition possesses significant practical implications. The variability and lack of structure in individuals’ handwriting pose asignificant obstacle for researchers in this field. This study aims to develop a system that can accurately identify handwritten Arabic characters. To achieve this, we employ two widely used techniques: convolutional neural networks (CNN) and support vector machines (SVM). Three models are proposed, one for each technique, as well as a hybrid model that combines the strengths of both approaches. The results are compared by selecting the optimal parameters for each technique. To assess the performance of the models, we use the MNIST digit database and the AHCD handwritten Arabic character database. The results show the superiority of the hybrid model with an accuracy of 98.21%.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"665 - 676"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098382","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Anomaly Detection in Surveillance Video: An Unsupervised Unity Encoder Approach","authors":"Manisha Sharma,&nbsp; Ravindra Kumar Purwar","doi":"10.3103/S1060992X26700086","DOIUrl":"10.3103/S1060992X26700086","url":null,"abstract":"<p>Video anomaly detection in crowded environments is a critical task in surveillance systems to ensure public safety. Traditional approaches relying on handcrafted features, statistical models and even traditional deep learning models often struggle with the complexity of real-world scenarios. To address these challenges, this paper proposes a unique encoder-decoder based anomaly detection framework that incorporates skip connections to preserve spatial features during reconstruction and Tanh activation in the bottleneck for a balanced latent space. The approach uses a Reconstruction Error that combines Mean Squared Error (MSE) and Standard Deviation (SD) to improve sensitivity to fine anomalies. The proposed framework is evaluated on benchmark datasets i.e. UCSD Ped1, Ped2, Avenue and Shanghai. It achieved a superior performance with higher AUC and reduced Equal Error Rate (EER) thus demonstrating its effectiveness in detecting anomalies in complex environments.</p>","PeriodicalId":721,"journal":{"name":"Optical Memory and Neural Networks","volume":"35 3","pages":"605 - 620"},"PeriodicalIF":0.9,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149098383","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}