Ship target detection of unmanned surface vehicle base on EfficientDet

Figures & data

Figure 1. Convolutional neural network structure.

Figure 2. Data flow graph of inverts residuals module.

Figure 3. Feature network design.

Figure 4. EfficientDet architecture.

Figure 5. The USV named Smart navigation I.

Figure 6. Dataset production.

Table

Algorithm 1. Code of GN
def GN (x, gamma, beta, G, eps=1e−5):
# x: input features with shape [B,C,H,W]
# gamma, beta: scale and offset, with shape [1,C,1,1]
# G: number of groups for GN
B, C, H, W = x.shape
x = tf.reshape(x, [B, G, C // G, H, W])
Mean,var = tf.nn.moments(x, [2, 3, 4], keep_dims = True)
x = (x−mean) / tf.sqrt(var + eps)
x = tf.reshape(x, [B, C, H, W])
return x*gamma + beta

Figure 7. Communication transmission settings.

Table 1. Running environment of model.

System Environment	Windows 10(64)
CPU	Intel(R)Core(TM) i9-9980XE
GPU	NVIDIA GTX2080Ti
Python	3.7.0
TensorFlow	2.2.0

Figure 8. The image of ship target detection from USV.

Table 2. Model comparison.

Model	FPS	Storage size (/MB)	AP50 (Simple)	AP50 (Complex)
EfficientDet-D4	21.4	83.5	0.936	0.875
EfficientDet-D3	34.5	47.9	0.824	0.764
EfficientDet-D0	42.5	16.1	0.792	0.650
Yolo V3	61	235.5	0.72	0.68
SSD-512	37.3	195.0	0.76	0.57
SSD-300	57.0	190.0	0.68	0.54
Faster-RCNN	39.2	119.0	0.66	0.57