95% accuracy achieved in quality detection for an agrochemical giant
The client is a known technology-based rice company from the US focused on developing smart rice with high yields and value-added traits. It carries out research to develop innovative agricultural products and efficient crop management solutions.
The client intended to automate the process of segregating seeds with hull in respective grades. It was keen on developing new-generation, efficient rice seeds with exceptionally high yield levels.
▪ The high resolution images of seeds with hulls captured using X-ray machine were manually inspected to correctly categorize them into respective grades.
▪ The existing manual nature of the process was highly time-consuming and very subjective, making it prone to human-error, which kept the grade classification accuracy low.
▪ Data was available, but incapability to address the data prevented the company from putting it to decision-making use.
Computer Vision, Python, GCP cloud services, Tensorflow
▪ Object detection models were used to interpret annotated images for correct insights.
▪ Feedback loops were provided to correct manual observations for wrong predictions.
▪ The corrected manual data was fed into algorithms to facilitate continuous improvement.
▪ The batch that would earlier take around 30 minutes, post the solution implementation executed the tasks in just 8 minutes, a staggering 73% reduction.
▪ The accuracy to predict existing 7 rice grades improved from 70% to 95%.
▪ The prediction model continuously enhances its accuracy with the flowing in of more and more data.