A Robust Approach of Detecting Sewer Cracks by Using Pixel Aggregation Network

In modern municipalities and infrastructure designing, the underground sewerage system is one of the most predominant parts to be cared about. It’s understood that, as the time passes and continuously usage of sewer pipes, the sewerage system susceptible to gradual deterioration which purely depends upon the degree of usage and location factors such as the passage of heavy traffic nearby or deep-rooted trees. Hence, this threat is having a huge probability to cause a massive financial and environmental disaster if it doesn’t monitor and detect. In order to do so and avoid the dreadful consequences, there is a substantial need for early detection and regular monitoring of these defects. To perform this task, CCTV inspection could be considered as an efficacious way of monitoring malfunction pipes. But, due to the extensive spreads of these pipes across several miles, so does the huge time consumption and tiring process makes the continuous inspection cumbersome and impotent. To overcome the limitations, make the process more systematic, and locate the defective places, we utilized the deep learning based technique. The strategy is based on Convolutional Neural Network (CNN), namely Pixel Aggregation Network (PAN) which mainly works on input data, properly labeled with deformable bounding boxes to train the model which further detects the arbitrary-shaped defects. Besides, in order to make it computationally effective, the segmentation head is utilized in PAN which optimizes the feature extraction along with keeping the computational cost at rock-bottom simultaneously. The data is comprising of 3150 images having distinctive crack samples along with 700 images for validation and testing. To validate the proposed method, the multiple metrics including Precision × Recall curve and Average Precision (AP) are exploited which gives the AP value of 79.71%. The qualitative, as well as quantities outcomes, show that the proposed method locates the malfunction area with high efficiency and accuracy.