Abstract
In recent years, convolutional neural networks (CNNs) have made great achievements in object extraction from very high-resolution (VHR) images. However, most existing approaches require large quantities of clean and accurate training data to achieve impressive classification results. The presence of inaccurate labels in training datasets is known to deteriorate the performance of CNNs. In this paper, we introduce a novel efficient method for improving the robustness when training CNN on the dataset with relatively noisy labels. First, we propose a feature and label noise model (FLNM) to model the noisy label distribution in the training dataset. Then, we use a multitask deep learning framework (MDLF) to integrate the FLNM into the training process of CNN. Finally, a novel loss function concerning the high-level features is introduced to efficiently train the MDLF. We evaluate our method on datasets from Massachusetts and compare this method with other state-of-the-art methods. The experimental results demonstrate the effectiveness of the proposed method in improving the classification performance of CNNs trained with noisy training dataset.
Highlights
Object extraction from very high-resolution (VHR) images is a very popular topic due to the wide range of applications for this technique, such as urban planning, land use analysis, disaster relief and automated map making
Our proposed method is conditioned on the image features but differs from these approaches in that our method explicitly models the relationship between the noisy labels and true labels
We propose a multitask deep learning framework (MDLF) that jointly learns to reduce the effects of the label noise and accurately classify the pixels in the images
Summary
Object extraction from very high-resolution (VHR) images is a very popular topic due to the wide range of applications for this technique, such as urban planning, land use analysis, disaster relief and automated map making. Veit et al [30] and Hendrycks et al [31] used an additional neural network to model the dependence between noisy labels and input images based on the information conferred by clean labels These methods require an additional accurately labeled dataset to obtain acceptable results. The model that comes closest to achieving the desired robustness is from Goldberger and Ben-Reuven [32], which directly learns the noise distribution from noisy labels without any clean labels These researchers added an extra layer before the classified layer to capture the dependence between the image features. A novel efficient method is proposed to train deep CNN models, aiming to produce robust image representations in the presence of noisy supervision for object extraction from VHR images.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.