Fig. 1. Results of image data enhancement. (a) Mixed image of rapeseed and weeds; (b) horizontally flipped image; (c) vertically flipped image; (d) horizontally and vertically flipped image; (e) brightness-enhanced image; (f) brightness-reduced image; (g) saturation-enhanced image; (h) saturation reduced image

Fig. 2. Framework of Faster R-CNN deep network

Fig. 3. Framework of RPN

Fig. 4. Comparison of total losses on SSD

Fig. 5. Comparison of total losses on faster R-CNN model

Fig. 6. Comparison of total losses for SSD andFaster R-CNN models

Fig. 7. Comparison of accuracy for two models

Fig. 8. Comparison of recall rates for two models

Fig. 9. Results of target recognition for rapeseed and weeds. (a) Results of recognition with occlusion; (b) results of recognition without occlusion; (c) results of recognition with complex background; (d) results of recognition with simple background

Table 1. Comparison of deep network models