Object detection is one of the critical and essential tasks in computer vision, with applications ranging from surveillance and industrial control to robotics and image analysis. This research presents a performance analysis of different YOLO (You Only Look Once) versions and a transformer model. The study evaluates YOLOv8, YOLOv9, YOLOv10, YOLOv11, and the RT-DETR (Real-Time Detection Transformer) model for object detection. The experiment uses a dataset of 1,730 images classified into five types: birds, dogs, cats, plants, and fruits, each with its subtypes. Also, we test with a frog dataset of 613 images which are characterized as complex images because present occlusion, complex backgrounds and variations in illumination. In addition, its performance is evaluated using standard metrics such as Precision, Recall, mAP50, and mAP50-95.

Computer vision methodologies using machine learning techniques usually consist of the following phases: pre-processing, segmentation, feature extraction, selection of relevant variables, classification, and evaluation. In this work, a methodology for object recognition is proposed. The methodology is called PSEV-BF (pre-segmentation and enhanced variables for bird features). PSEV-BF includes two new phases compared to the traditional computer vision methodologies, namely: pre-segmentation and enhancement of variables. Pre-segmentation is performed using the third version of YOLO (you only look once), a convolutional neural network (CNN) architecture designed for object detection. Additionally, a simulated annealing (SA) algorithm is proposed for the selection and enhancement of relevant variables. To test PSEV-BF, the repository commons object in Context (COCO) was used with images exhibiting uncontrolled environments. Finally, the APIoU metric (average precision intersection over union) is used as an evaluation benchmark to compare our methodology with standard configurations. The results show that PSEV-BF has the highest performance in all tests.

This work presents EDICA, a two-stage architecture for fine-grained image classification, which is a hybrid model for the detection and classification task. The model employs YOLOv8 for the detection stage and an ensemble deep learning model that utilizes a majority voting strategy for fine-grained image classification. The proposed model aims to enhance the precision of classification by integrating classification models that have been trained with the same classes. This approach enables the utilization of the strengths of these classification models for a range of test instances. The experiment involved a diverse set of classes, encompassing a variety of types, including dogs, cats, birds, fruits, frogs, and foliage; each class is divided into subclasses for finer-grained classification, such as specific dogs, cat breeds, bird species, and fruit types. The experimental results show that the hybrid model outperforms classification approaches that use only one model, thereby demonstrating greater robustness relating to ambiguous complex images and uncontrolled environments.