In recent years, advancements in digitalization and internet technologies have facilitated the widespread adoption of distance education and blended learning models in higher education institutions. In this process, Learning Management Systems (LMS) have become essential tools that enhance the quality of education by providing functions such as sharing educational materials, organizing exams, and monitoring student performance. Open-source LMS platforms stand out due to their customizable structures, cost advantages, and strong community support. This study analyzes the strengths and weaknesses of open-source LMSs commonly used in higher education institutions, such as Moodle, Open edX, Sakai, ILIAS, and Canvas. Based on criteria such as user-friendly interfaces, plugin options, data security, and technical support, this review assesses the systems' contributions to educational processes and user experiences. The findings reveal that Moodle and Open edX, with over 140 million and 14 million users respectively, excel in scalability and analytics capabilities, while platforms like Chamilo and Dokeos are more suitable for smaller institutions due to their simplicity and lightweight structure. The study concludes that open-source LMS platforms, with their flexible and economical structures, provide effective tools for supporting the digital transformation of education and offers guidance for selecting suitable platforms to meet institutional goals.

This paper presents a systematic and comprehensive review of deep learning (DL) methodologies used in precision agriculture (PA). It focuses on three critical application areas in particular: plant classification, plant disease detection, and weed detection. The study covers 93 peer-reviewed papers published between 2020 and 2025 and indexed in the SCI and SCI-Expanded indexed WoS database. Of these, 68 studies addressed disease detection, 13 focused on plant classification, and 12 examined weed detection strategies. The review describes a wide range of DL architectures, including Convolutional Neural Networks (CNNs), Residual Networks (ResNet), You Only Look Once (YOLO), Image Transformers (ViT), and various hybrid frameworks. A large number of models demonstrated exceptional performance with classification accuracies reaching up to 99.64% and precision and sensitivity values exceeding 98%. Studies have evaluated a wide range of datasets such as PlantVillage, COCO, and privately acquired RGB/UAV imagery, and a variety of sensor platforms such as drones, smartphones, hyperspectral, and LiDAR systems. Moreover, transfer learning and ensemble learning approaches have been frequently adopted to enhance generalization capabilities and model robustness. The integration of DL models with advanced technologies such as unmanned aerial vehicles (UAVs), unmanned ground robots (UGRs), depth-sensing cameras, and mobile-based platforms facilitates automation in agricultural monitoring, disease diagnosis, and yield prediction. This review not only consolidates the current technological developments, but also analyzes the emerging trends, methodological gaps, and possible directions for the advancement of sustainable, data-driven agricultural systems using artificial intelligence.

Digital search platforms enable real-time observation of relationship distress through behavioral traces. This study tests whether Google Trends predicts official divorce rates in the United States, Germany, the Netherlands, and the United Kingdom from 2009 to 2023. We introduce Digital Behavioral Decoupling, in which online distress signals diverge from legal outcomes as divorce shifts from an institutional procedure to an emotionally mediated digital phenomenon. Methods include unit root tests, cointegration analysis, Granger causality, spectral coherence, and rolling-origin nowcasting. Search queries are grouped into pre-divorce (cognitive distress), during-divorce (procedural action), and post-divorce (emotional recovery) phases. Results show 62.5% of terms lead divorce rates by 1–2 years, yet only 8.3% remain significant after False Discovery Rate correction. The Netherlands demonstrates 100% forecast improvements beyond autoregressive models (DM 2.87–3.43, p < 0.02) across all search terms, from early relationship therapy queries through procedural and post-divorce searches, indicating systematic capture of the entire divorce pathway. Germany shows intermediate results with 33% forecast success beyond autoregressive benchmarks (DM 2.40–2.81) limited to problem-recognition terms, suggesting episodic crisis-driven engagement. The United States and United Kingdom show no forecast gains beyond autoregressive models despite high search volumes, consistent with information saturation in normalized divorce cultures. Lead-lag relationships are frequency-specific, concentrated at 3–5 year periodicities. Findings link family sociology with affective computing and provide a replicable toolkit for tracking relationship dissolution in algorithmically curated information environments.

Çekişmeli Üretken Ağ (GAN), üretken bir derin öğrenme modeli olarak bilinir. Üretici (generator) ve ayırt edici (discriminator) yapılarından oluşmaktadır. Sentetik veri olarak bilinen GAN modeli çıktılarının oldukça başarılı örnekleri bilinmektedir. Farklı amaçlar ile kullanılabilen sentetik verilerin, başarılı bir şekilde üretilmesi durumunda insan gözü ile tespit edilebilmesi oldukça güç bir problemdir. Bu çalışmada farklı ve popüler Evrişimli Sinir Ağı (CNN) modellerinin öznitelik çıkarıcı olarak kullanıldığı, sentetik ve gerçek görüntüleri ayırt eden bu problem için Laplace filtresi ve benzemezlik tabanlı yeni bir CNN katmanı önerilmiştir. GAN modelinin farklı modeller üzerindeki başarı sonuçları tespit edilmiştir. Böylece, gözle ayırt edilemeyen sentetik verilerin tespiti için CNN modellerinden yararlanmanın uygun bir alternatif olduğu anlaşılmıştır. En iyi başarı %98.75 doğruluk oranıyla DenseNet ile elde edilmiştir.

Assessment of photovoltaic panel surfaces must cope with illumination variation, dust, snow, glare, cracks and bird droppings. This study presents SolarVisionNet (SVN), which fuses a seven-channel input (RGB plus physics-inspired glare, snow, bird-dropping and crack masks) with a 44-dimensional handcrafted descriptor through feature-wise linear modulation and adds an auxiliary head that predicts the masks. SVN is evaluated on a public six-class dataset under a leakage-aware protocol: because a naive random split places 38 of 131 test images next to a training near-duplicate, images are grouped into pseudo-sites, and results cover repeated seeds and grouped 5-fold cross-validation. The six-class model reaches 86.5 ± 3.3% accuracy and 86.8 ± 3.3% macro-F1, and the clean-versus-defective detector reaches 94.7% hold-out accuracy (90.4 ± 2.4% under cross-validation). In a nine-variant ablation, the physics-inspired components change macro-F1 by at most 2.7 points, within seed variability, and three of four ImageNet backbones trained with the same optimization recipe are equally accurate. Contrast-limited adaptive histogram equalization (CLAHE) lowers mean macro-F1 by 1.2 points over 50 paired runs, fixed-threshold rule overrides reduce macro-F1, and class-specific synthetic augmentation induces shortcut learning; the recommended configuration omits all three. SVN is a reproducible, leakage-audited baseline with inspectable glare, snow and bird-dropping masks.