Hose Failure Dataset

Table of Contents

[TOCM]

[TOC]

Dataset Description

This dataset was created specifically for the development of a real-time classifier using deep learning for image processing. The dataset contains images of failures in four types of hoses: SWAN, BICOLOR, GAS, and PRESION, captured from four ARDUCAM IMX519 cameras positioned at 90-degree intervals to analyze the perspective and integrity of the hoses in a 360-degree view. The images were collected at a resolution of 640x360 pixels.

The dataset currently has two versions: 1.0 and 2.0. Each version serves a different purpose The dataset is organized according to the following directory structure:

.
├── 1.0
│   ├── 1-1
│   │   ├── GAS_0
│   │   ├── GAS_1
│   │   ├── SWAN_0
│   │   └── SWAN_1
│   ├── 2-1
│   │   ├── BACKGROUND
│   │   ├── GAS_0
│   │   ├── GAS_1
│   │   ├── SWAN_0
│   │   └── SWAN_1
│   └── 4-1
│       ├── BACKGROUND
│       ├── GAS_0
│       ├── GAS_1
│       ├── SWAN_0
│       └── SWAN_1
└── 2.0
    └── 2-1
        ├── GAS
        │   ├── BACKGROUND
        │   ├── FAILURE
        │   ├── GOOD
        │   └── WARNING
        └── SWAN
            ├── BACKGROUND
            ├── FAILURE
            ├── GOOD
            └── WARNING

1.0

Version 1.0 is designed for binary classification between good and faulty hoses of the four types, considering three different perspectives 1, 2 , and 4 cameras per image. The directory structure for version 1.0 is as follows:

Details of dataset:

• Image size: 640 x 360 pixels
• Color space: RGB
• Number of classes: 4 (Background - Gas_0 - Gas_1 - Swan_0 - Swan_1)
• Number of samples per class: 500

---

• 1-1: One camera per image.
  • GAS_0: Images of GAS hoses (failure).
  • GAS_1: Images of GAS hoses (no failure).
  • SWAN_0: Images of SWAN hoses (failure).
  • SWAN_1: Images of SWAN hoses (no failure).

---

• 2-1: Two cameras per image.
  • BACKGROUND: Background images.
  • GAS_0: Images of GAS hoses (failure).
  • GAS_1: Images of GAS hoses (no failure).
  • SWAN_0: Images of SWAN hoses (failure).
  • SWAN_1: Images of SWAN hoses (no failure).

---

• 4-1: Four cameras per image.
  • BACKGROUND: Background images.
  • GAS_0: Images of GAS hoses (failure).
  • GAS_1: Images of GAS hoses (no failure).
  • SWAN_0: Images of SWAN hoses (failure).
  • SWAN_1: Images of SWAN hoses (no failure).

2.0

Details of dataset:

• Image size: 640 x 360 pixels
• Color space: RGB
• Number of classes: 4 (Background-Failure-Good-Warning)
• Number of samples per class: 1000

---

Version 2.0 of the dataset introduces additional categories and it focuses separately the failures for each type of hose. The directory structure for version 2.0 is as follows:

• 2-1: Two cameras per image.
  • GAS:
    • BACKGROUND: Background images.
    • FAILURE: Images of failed GAS hoses.
    • GOOD: Images of good GAS hoses.
    • WARNING: Images of warning signs on GAS hoses.
  • SWAN:
    • BACKGROUND: Background images.
    • FAILURE: Images of failed SWAN hoses.
    • GOOD: Images of good SWAN hoses.
    • WARNING: Images of warning signs on SWAN hoses.

Dataset Images Preview:

Version 1.0:

Here are preview images from each category, camera mode, and dataset version:

1-1:

GAS_0	GAS_1	SWAN_0	SWAN_1

2-1:

BACKGROUND	GAS_0	GAS_1	SWAN_0	SWAN_1

4-1:

BACKGROUND	GAS_0	GAS_1	SWAN_0	SWAN_1

---

Version 2.0:

2-1/GAS:

BACKGROUND	FAILURE	GOOD	WARNING

2-1/SWAN:

BACKGROUND	FAILURE	GOOD	WARNING

Repository Link

The code and additional information for the classifier development project using this dataset can be found in the following repository: Hose Failure Classifier