Tillage breaks the hardpan and loosens the soil, creating a fine, uniform seedbed that ultimately improves crop yield. This study aimed to develop a soil clod detection model and soil clod distribution maps after primary tillage, based on three clod-size categories: small (d < 100 mm), medium (100 mm ≤ d ≤ 250 mm), and large (d > 250 mm). An experiment was conducted at the research field at ICAR-CIAE with implemented geometry (IG) at three levels IG I (MB Plough), IG II (duck foot cultivator), and IG III (chisel cultivator), and two levels of moisture content (13.2 and 18.6%). GPS-tagged images were collected and annotated in ImageJ to determine clod parameters. The state-of-the-art deep learning-based object detection models (YOLOv7, YOLOv8, and YOLOv11) were used for automated clod classification. The models attained mean average precision values of 81.20%, 97.42%, and 95.44% respectively. The results revealed that clod size after primary tillage depends on the implement geometry and soil moisture. Computer vision-based clod-size distribution mapping can support IDSS frameworks by providing data for optimising secondary tillage operations, such as adjusting the u/v ratio of a rotavator to enhance soil pulverisation efficiency and tillage performance.
Computer vision-based soil clod distribution mapping after primary tillage
Okasha M.
;Kumar M.
2026-01-01
Abstract
Tillage breaks the hardpan and loosens the soil, creating a fine, uniform seedbed that ultimately improves crop yield. This study aimed to develop a soil clod detection model and soil clod distribution maps after primary tillage, based on three clod-size categories: small (d < 100 mm), medium (100 mm ≤ d ≤ 250 mm), and large (d > 250 mm). An experiment was conducted at the research field at ICAR-CIAE with implemented geometry (IG) at three levels IG I (MB Plough), IG II (duck foot cultivator), and IG III (chisel cultivator), and two levels of moisture content (13.2 and 18.6%). GPS-tagged images were collected and annotated in ImageJ to determine clod parameters. The state-of-the-art deep learning-based object detection models (YOLOv7, YOLOv8, and YOLOv11) were used for automated clod classification. The models attained mean average precision values of 81.20%, 97.42%, and 95.44% respectively. The results revealed that clod size after primary tillage depends on the implement geometry and soil moisture. Computer vision-based clod-size distribution mapping can support IDSS frameworks by providing data for optimising secondary tillage operations, such as adjusting the u/v ratio of a rotavator to enhance soil pulverisation efficiency and tillage performance.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
