The Impact of ResNet50 on Image Recognition Accuracy Compared to Prior Convolutional Neural Networks

Abstract

This research evaluated the impact of the ResNet50 convolutional neural network architecture on improving image classification accuracy. Questionnaire data was collected from 386 deep learning experts regarding the accuracy, efficiency, and real-world performance of ResNet50 versus earlier CNNs like VGG16, InceptionV3, and AlexNet. Quantitative analysis showed ResNet50 achieves significantly higher accuracy, with gains of 3-4% on average based on respondent ratings. Qualitative feedback indicated ResNet50’s innovative residual learning approach enables training much deeper networks, allowing more sophisticated feature extraction. Critically, ResNet50 attains superior accuracy with fewer parameters than previous models. Statistical tests confirmed ResNet50 demonstrates significant improvements in accuracy, efficiency, and applicability over earlier CNNs due to its advanced deep residual design. The results provide strong evidence that ResNet50’s architectural innovations substantially enhance image classification accuracy and real-world computer vision applications.