Erdinç Avaroğlu

Oyster mushroom cultivation requires accurate object detection for automated monitoring and precision agriculture applications. This study evaluates a structure-aware three-channel input representation for YOLOv8s in which raw RGB channels are replaced by grayscale intensity, Sobel gradient magnitude, and a complementary structural channel derived from Gaussian Blur, Laplacian of Gaussian (LoG), Canny edge detection, or Gabor filtering. The dataset contains 555 RGB images and 8282 maturity-labeled mushroom instances. Controlled experiments include an RGB baseline with HSV augmentation disabled, component-wise ablations (GGG and GGradG), repeated-seed training, a chronological holdout, and a cross-architecture RT-DETR evaluation. Under the fixed random split, differences among RGB, grayscale, and structure-aware inputs were modest, and the ablations indicate that grayscale conversion accounts for most of the measured effect. Performance decreased substantially under the chronological split, and RT-DETR did not reproduce the same ordering observed with YOLOv8s. These results show that input representation can influence detector behavior, but they do not support a general claim that handcrafted structural channels consistently improve robustness across evaluation protocols or architectures.

Accurate localization of bounding boxes is a prerequisite for enabling vision-driven precision agriculture pipelines, as downstream tasks such as morphological feature extraction, growth monitoring, and digital-twin synchronization depend directly on the geometric quality of the detected boxes. Distance-IoU (DIoU) and Complete-IoU (CIoU) improve upon simple overlap-based objectives by incorporating a normalized center-distance term into the regression loss, along with the overlap and, for CIoU, an aspect-ratio penalty, but that term remains embedded in a single composite formulation with an implicit, non-adjustable weight. We propose a Directional Alignment Penalty (DAP), an auxiliary localization regularizer that introduces an independently weighted normalized center-displacement term into the bounding-box regression objective without modifying the detector architecture. The proposed DAP-YOLOv8 increased mAP@0.5:0.95 from 51.69% to 53.76% and mAP@0.5 from 80.03% to 80.36% while preserving precision and recall; repeated-seed experiments further showed that the improvement in fine-grained localization was consistent across random initializations on a purpose-built oyster-mushroom (Pleurotus ostreatus) dataset collected from a real-world smart greenhouse. Results show that explicitly modeling the center-distance factor as an independent and tunable component can improve fine-grained localization without sacrificing the computational efficiency of the base detector, thereby providing a lightweight plug-in extension for agricultural detection and digital-twin applications.

We present a publicly reproducible, multi-source behavioral-anomaly benchmark for insider-threat detection on theLANL cyber1 corpus. Our contribution is deliberately methodological and empirical—a leakage-aware engineer-ing pipeline, a labeled red-team evaluation, multi-source feature fusion, and a public reproduction package—ratherthan a new detection algorithm. Insider-threat signals in enterprise authentication logs matter for security opera-tions, yet public corpora with aligned incident labels remain uncommon. We study window-level behavioral anomalydetection on the LANL cyber1 release, a de-identified enterprise timeline paired with red-team event tags, wherepositives are extremely rare and operational triage must be budgeted explicitly. We construct past-aware, day-lengthtabular features (including past-only rarity to mitigate look-ahead) and compare Isolation Forest, one-class SVM, areconstruction autoencoder, and an LSTM trajectory autoencoder, then extend to multi-source fusion of authentica-tion, process telemetry, and flow-derived aggregates. Besides standard ranking summaries, we foreground recall atthe 80th percentile of unlabeled scores as a workload-oriented operating point. On the fused grid, Isolation Forestimproves AUC-ROC from 0.7707 (authentication only) to 0.8719 with process and flows—about +0.10 absolute—raising Recall@p80 from 56.8% to 76.7%; five-fold cross-validation yields AUC-ROC 0.8664±0.0295. By contrast,the LSTM model is neutral-to-mixed, and naive weighted-z late fusion is an honest negative, reducing AUC-ROCto 0.7473. Reproducible engineering, complementary feature fusion, and workload-facing operating points outweighdefault model depth when labels and positives are extremely sparse.

Object detection in remote sensing images is still recognized as a demanding task, largely because of significant scale differences among objects and the complexity of background scenes. Detecting small and medium-sized targets within cluttered environments, in particular, continues to challenge many existing algorithms. To address these issues, this study presents a new model named AERIS-ED (Attention-Enhanced Real-time Intelligence System for Efficient Detection). The framework adopts a C3 (Cross Stage Partial with three convolutions) based backbone and incorporates Efficient Attention (EA) units, but unlike conventional designs, these modules are inserted only at the P3 and P4 levels of the feature pyramid. This focused integration enables richer feature interaction across scales and enhances the recognition of small and medium objects. Comprehensive experiments on the MAR20 and VEDAI datasets highlight the benefits of the proposed approach. On MAR20, the model achieves a mean Average Precision at an Intersection over Union threshold of 0.5 (mAP@0.5) of 95.1% with an inference latency of only 3.8 ms/img. On VEDAI, it secures 83.0% mAP@0.5 while maintaining the same efficiency, thereby confirming its suitability for real-time applications. Overall, the results indicate that AERIS-ED strengthens detection accuracy for small objects without compromising computational speed. These improvements suggest that the architecture is not only promising for multi-scale detection research but also has strong potential in practical remote sensing tasks.

Underwater images are affected by various issues such as insufficient illumination, low contrast, blur, and color distortion. These distortions significantly degrade the performance of underwater imaging systems and can lead to functional failures in some applications. The primary objective of this work is to obtain information-rich, high-quality images by eliminating distortions caused by the underwater environment. The proposed approach first focuses on achieving realistic tones by removing color distortions in the image through color correction using a gray-world assumption. Then, overexposed and underexposed images are generated using a Gaussian-filtered luminance map. These various variations are combined with a powerful unsupervised deep learning-based fusion algorithm to obtain the final enhanced image. Comprehensive visual and objective experiments were conducted to demonstrate the method's performance. In comparisons conducted using objective metrics that evaluate image quality from different perspectives, the proposed method outperformed its competitors; in particular, it outperformed its closest competitor, UIQM, which evaluates perceptual quality, color, and contrast together, by approximately 6.3%. Additionally, visual comparisons demonstrate that the method is more effective at revealing subtle structural details and background elements.

The underwater environment severely degrades image quality by absorbing and scattering light. This causes significant challenges, including non-uniform illumination, low contrast, color distortion, and blurring. These degradations compromise the performance of critical underwater applications, including water quality monitoring, object detection, and identification. To address these issues, this study proposes a detail-oriented hybrid framework for underwater image enhancement that synergizes the strengths of traditional image processing with the powerful feature extraction capabilities of unsupervised deep learning. Our framework introduces a novel multi-scale detail enhancement unit to accentuate structural information, followed by a Latent Low-Rank Representation (LatLRR)-based simplification step. This unique combination effectively suppresses common artifacts like oversharpening, spurious edges, and noise by decomposing the image into meaningful subspaces. The principal structural features are then optimally combined with a gamma-corrected luminance channel using an unsupervised MU-Fusion network, achieving a balanced optimization of both global contrast and local details. The experimental results on the challenging Test-C60 and OceanDark datasets demonstrate that our method consistently outperforms state-of-the-art fusion-based approaches, achieving average improvements of 7.5% in UIQM, 6% in IL-NIQE, and 3% in AG. Wilcoxon signed-rank tests confirm that these performance gains are statistically significant (p < 0.01). Consequently, the proposed method significantly mitigates prevalent issues such as color aberration, detail loss, and artificial haze, which are frequently encountered in existing techniques.

Döner kanatlı insansız hava araçları (İHA), çeşitli uygulamalarda sağladıkları esneklik nedeniyle dikkat çekmektedir. Çeşitli alanlarda, geniş bir yelpazede kullanılan döner kanatlı İHA’ların daha fazla yaygınlaşmasının önündeki en büyük engel, bataryalarının kısa sürede deşarj olmasından dolayı havada kalma sürelerinin kısa olması olarak görülmektedir. Döner kanat İHA’lar da kullanılan lityum polimer (Lipo) bataryaların bir diğer dezavantajlı durumu ise kullanım ömrüdür. Lipo bataryaların toplam verebildiği akım miktarının bir göstergesi olan pil seviyesi sürekli kontrol edilmemesi ve bu seviyenin 20%’nin altına düşüşü pil ömrünü kısaltmakta ve zaman zaman yapısını tamamen bozarak İHA’yı çeşitli kırımlara uğratabilmektedir. Özellikle otonom olarak sürekli görevlerde kesintisiz çalıştırılmak istenen döner kanatlı İHA'ların; pil ömrünün uzatılması ve belirli bir pil seviyesine indiğinde otonom olarak en yakın şarj istasyonuna güvenli inişleri önem taşımaktadır. Bu çerçevede yapılan çalışmada, belirlenen bir güzergah boyunca otonom uçan bir İHA'nın iniş platformuna yatay olarak yaklaşırken ve bu platforma dikey iniş yaparken batarya seviyesindeki azalma miktarı, makine öğrenmesi algoritmaları kullanılarak tahmin edilmiştir. Uçuş sonunda, İHA'nın güvenli bir şekilde istenilen batarya seviyesinde yere iniş yapması hedeflenmiştir. Yatay seyir sırasında, güzergah noktaları üzerinde anlık veriler kullanılarak tahminler yapılmıştır. Dikey iniş esnasında ise, görüntü işleme teknikleri kullanılarak, farklı irtifalardan indirilen İHA'dan elde edilen verilerle tahminler gerçekleştirilmiştir. Çalışma kapsamında tasarlanan İHA ile gerçek saha koşullarında ve farklı irtifalarda testler gerçekleştirilmiştir. Kalkış öncesinde, İHA’ya belirli bir batarya seviyesi bildirilmiş ve uçuşa başlanmıştır. Uçuş sonrasında ise İHA’nın batarya seviyesi kontrol edilerek başlangıç seviyesi ile arasındaki farklar gözlemlenmiştir. Yapılan testlerde, en yüksek 3%, en düşük ise 0% batarya seviyesi farkına ulaşılmıştır.Bu çerçevede yapılan çalışmada, belirlenen bir güzergah boyunca otonom uçan bir İHA'nın iniş platformuna yatay olarak yaklaşırken ve bu platforma dikey iniş yaparken batarya seviyesindeki azalma miktarı, makine öğrenmesi algoritmaları kullanılarak tahmin edilmiştir. Uçuş sonunda, İHA'nın güvenli bir şekilde istenilen batarya seviyesinde yere iniş yapması hedeflenmiştir. Yatay seyir sırasında, güzergah noktaları üzerinde anlık veriler kullanılarak tahminler yapılmıştır. Dikey iniş esnasında ise, görüntü işleme teknikleri kullanılarak, farklı irtifalardan indirilen İHA'dan elde edilen verilerle tahminler gerçekleştirilmiştir. Çalışma kapsamında tasarlanan İHA ile gerçek saha koşullarında ve farklı irtifalarda testler gerçekleştirilmiştir. Kalkış öncesinde, İHA’ya belirli bir batarya seviyesi bildirilmiş ve uçuşa başlanmıştır. Uçuş sonrasında ise İHA’nın batarya seviyesi kontrol edilerek başlangıç seviyesi ile arasındaki farklar gözlemlenmiştir. Yapılan testlerde, en yüksek 3%, en düşük ise 0% batarya seviyesi farkına ulaşılmıştır.

Assessing the quality of agricultural products holds vital significance in enhancing production efficiency and market viability. The adoption of artificial intelligence (AI) has notably surged for this purpose, employing deep learning and machine learning techniques to process and classify agricultural product images, adhering to defined standards. This study focuses on the lemon dataset, encompassing ‘good’ and ‘bad’ quality classes, initiate by augmenting data through rescaling, random zoom, flip, and rotation methods. Subsequently, employing eight diverse deep learning approaches and two transformer methods for classification, the study culminated in the ViT method achieving an unprecedented 99.84% accuracy, 99.95% recall, and 99.66% precision, marking the highest accuracy documented. These findings strongly advocate for the efficacy of the ViT method in successfully classifying lemon quality, spotlighting its potential impact on agricultural quality assessment.