Good and Bad Seeds Detection Using Image Processing and Deep Learning with Automatic Sorting

Seed quality decides how well a crop germinates, how uniform the field is and how much it yields, yet most seed lots are still graded by eye, which is slow, tiring and different from one person to the next. This paper presents the design of a low-cost system that detects good and bad seeds from images and sorts them automatically. A camera inside a lighting box captures each seed, image processing separates the seed from the background, and a lightweight convolutional neural network (CNN) classifies it as good or bad. A Raspberry Pi drives a servo gate that diverts the seed to the correct bin, shows the result on an LCD and publishes the counts over MQTT so that a phone shows the running totals. A decision rule with a confidence band sends doubtful seeds to the reject bin. The paper gives the processing pipeline, a dataset plan, evaluation metrics, a latency budget and a test protocol. No measured results are reported: the figures in the paper are design targets, and the result tables are laid out for the measurements on the prototype, which are planned as the next stage