Computer-Vision-Based Fruit Grading System Increases Throughput 6-Fold
Published on November 2, 2023 by TIS Marketing.
Traditionally, orchard owners have relied on manual labor for the grading and sorting of citrus fruit. The task is labor-intensive and tedious for humans which results in mistakes and reduced throughput. When faced with this problem, an orchard owner with a rapidly growing business in Kaohsiung, Taiwan partnered with Hitspectra, an intelligent inspection and vision system integrator. Hitspectra helped the citrus grower rapidly increase inspection and sorting efficiency with an innovative computer-vision-based inspection system. Using software that they developed in-house, Hitspectra integrated The Imaging Source's 5 MP USB 3 cameras into the orchard's existing sorting system to create a robust inspection and grading system that has increased overall efficiency by 6-fold.
Deep Learning Approach Increases Inspection ScopeEarly automated fruit sorting used rule-based inspection to grade, for example, fruit color and size. Deep learning methodologies have developed to enable vision solutions for applications whose inspection criteria are not readily quantified or defined, allowing for a wider range of important inspection and sorting criteria: blemishes, rot, mold, and other defects. Hitspectra was able to integrate the new inspection station into an existing weight screening machine at the orchard's sorting facility. The industrial color cameras capture images from multiple angels of each lemon as it proceeds along the conveyor. The image data is then segmented, fused, and analyzed against the training data for shape, size, color, and surface defects. Based on the training data, the system then sorts the lemons as "OK" or "NOK".
Vision System Maximizes Throughput With Less LaborPrior to the integration of Hitspectra's vision solution, three employees processed around 400 kilograms of lemons per hour. The new system allows one employee to process 800 kilograms per hour, increasing throughput by an astounding 6-fold. The accuracy rate of the AI-based system has increased to 90% and Hitspectra expects to see additional improvements after additional updates and training data. Hitspectra's manager commented: "With the successful introduction of automated visual inspection [in agricultural applications], manual labor can be reduced, and the inspection screening criteria can be quantified and standardized."