Дослідження Методів Формування Двохградаційного Коду Зображень Зернового Потоку

Introduction. Detecting individual seeds in an unordered grain flow in real time requires reliable conversion of optical signals into compact binary images. The study addresses the loss of object-boundary information caused by conventional single-threshold quantization. Materials and Methods. Video signals from a scanning optoelectronic sensor were considered. A two-threshold quantization method was analyzed, in which the binary state of a pixel depends on signal amplitude and on the state of the corresponding pixel in the preceding scan line. Results. A hysteresis-based decision rule for constructing one-bit images of grains was proposed. It preserves boundary pixels at intermediate shading levels when their counterparts were detected in the previous scan line. A functional implementation employing threshold comparators, multiplexers, logic and memory is considered. In archival bench tests, the complete seed-counting system was assessed in 103 trials; a separate simulation on 200 synthetic arrays yielded F1 scores of 93.26% (hysteresis) and 91.14% (single threshold) at 0.9 V noise. Conclusions. Two-threshold hysteresis quantization makes it possible to obtain informative binary images while keeping one-bit encoding. The approach can simplify registration and subsequent processing of an unordered grain flow.