A Deep Learning Based Approach to Child Labour Detection

Currently, the world is facing extreme turmoil due to the spread of the COVID-19 virus, but the effects of this detrimental virus have been exceedingly dreadful for developing countries like Bangladesh. Being a developing country, Bangladesh has many social concerns to overcome and among them child labour has been a constant one for decades. It has been anticipated by several researchers that the contingency of COVID-19 virus would cause a surplus of such child labour occurrences throughout the world especially in the South Asian region. Considering the concern, we are proposing a benevolent yet technical solution that has the potential to eradicate the problem on a large scale. This paper beholds the idea of a child labour detection technique to prevent such illegal instances taking place in the construction areas, brickfields, coal mining fields, make-up product mining fields, etc. on the South Asian countries like Bangladesh, India, Pakistan, Nepal. Considering the unavailability of an open sourced child labourers dataset in the context of the mentioned region, a noble dataset of such kind has been formulated in this project. A child labour recognizing model has been built based on Convolutional Neural Network (CNN) using the dataset. Additionally, the resulting accuracy has also been compared using two well-known transfer learning methods for better perception. The CNN model possessed 90.625% test accuracy whereas the two transfer learning based models showed 95.312% and 96.875% respectively.