基于迁移学习的图像识别案例分析
Case Analysis of Image Recognition Based on Transfer Learning
庞德鹏1
作者信息
摘要
阐述基于深度学习的皮肤癌识别在精度上不断取得突破,但是受限于医疗图像数据获取困难且成本高昂,皮肤癌识别的大规模实际应用仍充满挑战.为分析迁移学习对皮肤癌识别任务的提升效果,选择经典CNN模型,迁移网络模型EfficientNet V2和Vision Transform比较其性能.结果表明,在模型推理时间和推理性能权衡方面,EfficientNetV2模型表现超越其他模型,更适合在现实情况中使用.
Abstract
This paper describes the continuous breakthroughs in accuracy in skin cancer recognition based on deep learning.However,due to the difficulty and high cost of obtaining medical image data,the large-scale practical application of skin cancer recognition is still full of challenges.To analyze the improvement effect of transfer learning on skin cancer recognition tasks,classic CNN models,EfficientNet V2 and Vision Transform were selected to compare their performance.The results indicate that in terms of balancing model inference time and inference performance,EfficientNet V2 model outperforms other models and is more suitable for use in real-world situations.
关键词
智能算法/深度学习/CNN模型/图像识别Key words
intelligent algorithms/deep learning/CNN models/image recognition引用本文复制引用
出版年
2024
NETL
NSTL
万方数据