Seeding detection and distribution evaluation using the developed automatic maize seeding machine

Abstract

Seeding is a time-consuming and labour-intensive process in the detection of maize seeds germination rate. Automatic seeding can reduce labour requirements and improve efficiency. The quality of automatic seeding plays an important role in collecting accurate germination rates. In this research, a newly developed automatic maize seeding machine (AMSM) was used to detect the maize seeds and evaluate distribution by advanced computer vision and analysis methodologies. Three methods were utilized to detect maize seeds after they were loaded into the holes of the seed plate of AMSM: colour segmentation + connected component analysis, support vector machine (SVM) + sliding window, and YOLOV5. The results showed that compared with colour + connected component analysis and sliding window + SVM, YOLOV5 exhibited higher accuracy and improved computational efficiency. When the input image pixel size was 960 × 960, the precision of YOLOV5 + merge-NMS was 97.3 %, the recall was 0.918, the F1-score was 0.945, and the processing time of a single image was 0.38 ms, which is the optimal performance. YOLOV5 + merge-NMS demonstrated robust detection maize seeds in images with different illumination equalization processing. Based on maize seeds detection results, seeding distribution evaluation was realized with an average accuracy of 91.14 %. In addition, the normal seeding of the AMSM was 82.3 %. Utilization of YOLOV5 + merge-NMS enables rapid and precise detection of maize seeding while establishing a basis for accurately calculating seed distribution. The AMSM can meet the requirements of automatic maize seeds detection and distribution evaluation and the developed methodologies have the potential for automatic germination detection.