An Iteratively Optimized Patch Label Inference Network for Automatic Pavement Distress Detection


We present a novel Deep Learning framework that we call the Iteratively Optimized Patch Label Inference Network (IOPLIN). This network is designed to automatically detect a wide variety of pavement distresses, not just specific ones like cracks and potholes. IOPLIN can be trained iteratively using only the image label by employing the Expectation-Maximization Inspired Patch Label Distillation (EMIPLD) strategy. It is able to successfully complete this task by deducing the labels of patches from images of pavement. IOPLIN has a number of advantageous advantages over other state-of-the-art single branch CNN models like GoogleNet and EfficientNet. These advantages include: Since IOPLIN extracts the visual features from unrevised image patches rather than the resized entire image, it is able to handle images of varying resolutions and adequately utilize image information, particularly for high-resolution images. This is made possible by the fact that IOPLIN does not resize the entire image. In addition to this, it is able to roughly localize the pavement distress without making use of any prior localization information during the training phase. We construct a large-scale Bituminous Pavement Disease Detection dataset that we call CQU-BPDD. This dataset is comprised of 60,059 high-resolution pavement images that are acquired from different locations and at different times. The purpose of this is to better evaluate how effective our method is when it is put into practice. Extensive results on this dataset demonstrate that IOPLIN is superior to the state-of-the-art image classification approaches in automatic pavement distress detection. These results were obtained by comparing IOPLIN's results to those of other methods. IOPLIN's source codes are available for download at, and the CQU-BPDD dataset can be viewed at You can find more information about both of these resources on DearCaat's website.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here

PROJECT TITLE : Accelerating GMM-Based Patch Priors for Image Restoration Three Ingredients for a 100_ Speed-Up ABSTRACT: The goal of picture restoration is to restore a clear image from a smudged one. In order to restore natural
PROJECT TITLE : High-quality Image Restoration Using Low-Rank Patch Regularization and Global Structure Sparsity ABSTRACT: In recent years, picture restoration has improved significantly thanks to techniques based on nonlocal
PROJECT TITLE : Nonlocal Patch Tensor Sparse Representation for Hyperspectral Image Super-Resolution ABSTRACT: Super-resolution of hyperspectral images (HSI) is shown in this study, which combines low-resolution (LR) HSI with
PROJECT TITLE :A Group-Based Image Inpainting Using Patch Refinement in MRF Framework - 2018ABSTRACT:This Project presents a Markov random field (MRF)-primarily based image inpainting algorithm using patch selection from teams
PROJECT TITLE :Robust Multi-Exposure Image Fusion: A Structural Patch Decomposition Approach - 2017ABSTRACT:We tend to propose a easy however effective structural patch decomposition approach for multi-exposure image fusion (MEF)

Ready to Complete Your Academic MTech Project Work In Affordable Price ?

Project Enquiry