Author Affiliations
School of Information Engineering, Inner Mongolia University of Science and Technology, Baotou, Inner Mongolia 014010, Chinashow less
Fig. 1. Schematic diagram of the technical solution
Fig. 2. Structure diagram of the YOLOv5s model
Fig. 3. Schematic diagram of the angle between feature parts
Fig. 4. Structure of the fully connected neural network
Fig. 5. Labeling of feature parts
Fig. 6. Training results of the YOLOv5s model. (a) Loss curve; (b) precision; (c) recall; (d) mAP
Fig. 7. Detection results of cattle feature parts
Fig. 8. Sample image of the cattle behavior. (a) Standing behavior; (b) lying behavior; (c) feeding behavior
Fig. 9. Training curve of the fully connected neural network model. (a) Loss curve; (b) accuracy curve
No. | Characteristics value meaning |
---|
1--3 | width of the target box of cattle, body and head (cattle_w, body_w, head_w) | 4--6 | height of the target box of cattle, body and head (cattle_h, body_h, head_h) | 7 | distance from head to tail (headtotail_dis) | 8--18 | distance from head to joint, knee and hoof (headtojoint_dis1-3, headtoknee_dis1-4, headtohoof_dis1-4) | 19--29 | distance from tail to joint, knee and hoof (tailtojoint_dis1-3, tailtoknee_dis1-4, tailtohoof_dis1-4) | 30--40 | angle between the head and joint, knee, and hoof relative to the tail (head-tail-joint_ang1-3, head-tail-knee_ang1-4, head-tail-hoof_ang1-4) | 41--51 | angle between the tail and joint, knee and hoof relative to the head (tail- head-joint_ang1-3, tail- head-knee_ang1-4, tail- head-hoof_ang1-4) |
|
Table 1. Vector format of feature part spatial relationship
Recognition category | mAP | Cattle | Body | Head | Tail | Joint | Hoof | Knee |
---|
AP | 90.9 | 94.0 | 94.9 | 97.2 | 87.7 | 84.4 | 91.1 | 87.2 |
|
Table 2. Training accuracy of the YOLOv5s model unit: %
Standardized processing | Dropout processing | All behavior | Standing | Lying | Feeding |
---|
× | × | 96.2 | 96.8 | 96.2 | 95.3 | × | √ | 96.7 | 96.4 | 98.7 | 90.6 | √ | × | 97.2 | 97.2 | 98.7 | 95.3 | √ | √ | 97.7 | 97.6 | 100 | 95.3 |
|
Table 3. Classification accuracy of fully connected neural network model unit: %
Standardized processing | All behavior | Standing | Lying | Feeding |
---|
× | 91.9 | 91.7 | 93.6 | 90.6 | √ | 96.7 | 97.6 | 96.2 | 93.8 |
|
Table 4. Classification accuracy of decision tree model unit: %
Behavior | Standing | Lying | Feeding |
---|
Real time /s | 232.1 | 262.4 | 88.3 | Predicted time /s | 238.0 | 262.0 | 83.0 | Relative error /% | 2.54 | -0.15 | -6.00 |
|
Table 5. Statistics of cattle behavior time