Helping a Waste Management Company Enhance Recycling Efforts with Image Segmentation

Services Provided: Image Annotation

Our customer is a waste collector/management company who was working on an AI solution that could detect what kind of waste is being disposed of in the containers of businesses (their customers). Through the implementation of image annotation and computer vision, the tool can recognize what exactly is in the customers’ containers: the type of waste,  its classes, the number of items, etc. 

The AI solution is applied to track how well the companies’ recycling efforts are going and make sure that the waste sorting rules are kept to ensure safety. For example, if it appears that there are dangerous products in the waste container of a customer, the company can send a warning or a fine to this client. Also, the tool provides more visibility into waste management processes and ensures the quality of the recycled materials, so the company can create better products from the collected waste.

The client needed to annotate images containing waste materials and define certain types of trash. This includes items like paper, hard plastic, wood, garbage bags, glass, food, and dangerous waste. All objects that fall under these categories have to be annotated with a polygon and the relative class has to be applied. Image segmentation requires a high level of training and attention to detail on behalf of the data labeling team. The main idea was to define the type of waste correctly in a specific client’s tool.

The company reached out to Mindy Support about data annotation service due to the recommendation of their partner, which is our long-standing client. In addition, the company previously had an experience of outsourcing to Mindy Support its data sorting project and it was happy with the results. Therefore, the company decided to trust us with its new project, but this time they needed image annotation work to be done.

Mindy Support assembled a team of data annotators to perform the annotation work. We have extensive experience with data labeling and similar projects in our portfolio, so our team did not need a whole lot of training. We labeled all of the various types of waste in the images, reaching a quality score of 98% after the annotation phase without any additional quality validation step. This exceeded the client’s expectations by +3%. The client was very satisfied with the results and decided to continue the project as well as possibly adding more classes to the taxonomy in the near future to meet the needs of more advanced recycling technology.

• Achieved a quality score of 98%, which exceeded the client’s expectations by +3%
• Successful cooperation for more than 2 years