Application and Performance Evaluation of Object Detection Technology Based on Computer Vision

In light of the challenges posed by the variability of industrial part shapes, weak texture information, and surface reflections in current industrial vision inspection, traditional methods such as feature descriptors and template matching are inadequate. This paper innovatively proposes a machine learning object detection method, establishes robust stable features for weakly textured industrial parts, and evaluates the performance of the model and feature selection in combination with actual fuel tank cap detection tasks. Firstly, we construct a dataset of fuel tank cap targets and establish a comprehensive and relatively stable feature dataset, including geometric features, texture features, and color features. Secondly, through ablation experiments and comparative experiments, we conduct intelligent per-formance testing on model detection accuracy, parameter selection, and feature contribution. Finally, the evaluation results demon-strate that the support vector machine (SVM) model exhibits the best detection performance after feature selection, with a detection accuracy of 99.0% under the optimal parameter settings validated through experiments.