The intelligent fruit sorting line designed and built by Suzhou Jiding Network Technology has completed final acceptance and entered production. The system grades fruit quality automatically using hyperspectral imaging and multi-sensor fusion, and sorts it with flexible robots — raising throughput while cutting the physical damage caused by manual handling.
Project background: from human eyes to a model
Before the project the plant graded entirely by hand. Standards differed between shifts, peak season outpaced capacity, and every additional manual pick added bruising that pushed shrink up. After go-live, grading is done by the hyperspectral imaging + AI model, sorting is executed by flexible robots following the grade result, and management reads throughput, grade mix and shrink from a live dashboard.
Two deliberate trade-offs
- Grading decoupled from control. The AI grading engine is deployed independently: changing a camera model or retraining the model later does not require touching the line control software, which lowers both upgrade cost and downtime risk.
- Damage-free before maximum cycle time. Robot pick paths treat "protect every piece of fruit" as the first constraint, trading a small amount of cycle time for a lower bruising rate — shrink is the client's most direct cost.
As agreed with the client, this page does not show data that could identify them. De-identified improvement figures are available during pre-contract discussions, by industry.
What it means for similar projects
If you are evaluating software for a grading or inspection step, bring these three questions: is the current bottleneck judgement, speed or missing data? where is the responsibility boundary between the equipment vendor and the software team? and who owns model retraining, at what price? The answers largely decide whether the project succeeds.