Research Article

Underwater Image Processing and Object Detection Based on Deep CNN Method

Table 1

mAP and precision of different iteration times by Fast RCNN, Faster RCNN, and YOLO V3. ().

IterationFast RCNNFaster RCNNYOLO V3
mAP (%)Precision (%)mAPPrecision (%)mAP (%)Precision (%)
Sea cucumberSea urchinScallopSea cucumberSea urchinScallopSea cucumberSea urchinScallop

200027.2630.1326.7924.8727.5330.1827.2925.1335.4337.1435.4233.74
400037.5640.5138.2333.9338.7440.8039.3536.0645.9048.1245.5044.08
600041.8344.4541.3639.6743.1545.3042.8041.3549.6151.8149.8747.16
800045.3748.6745.8541.5946.5948.3547.3544.0852.4054.5653.1449.51
1000048.2251.3347.8445.5050.2852.0950.7648.0055.8958.1756.9952.50
1200050.9053.7551.3147.6552.5353.9653.4450.2058.3459.7759.4855.77
1400053.0955.6954.2049.3854.4356.1854.7852.3160.5863.3960.9157.44
1600055.0458.8554.9251.3557.3259.6656.9855.3462.0264.0962.5059.47
1800056.6660.4956.8152.6758.6260.3559.5555.9564.1866.7965.1560.58
2000058.6362.1258.4955.2760.9362.3061.8658.6366.0068.8766.2062.93
2200060.4263.9560.6356.6763.0764.3363.9360.9567.2270.3768.0263.26
2400061.3564.5762.1957.2964.3765.9464.6762.5168.8871.5070.5464.60
2600063.4066.6063.9459.6566.3868.0767.1763.9070.4472.8571.8366.64
2800065.0368.8165.3660.9268.1569.9868.6765.8272.0074.7973.1768.03
3000066.8470.0968.1962.2469.4270.4970.5367.2471.9974.8473.4467.70
3200067.6870.7368.5363.7870.6872.7871.4767.7972.2474.4773.7868.47
3400069.2672.6571.0364.1171.7273.9672.1769.0172.1574.6274.0167.81
3600070.9674.7572.2365.9073.7574.4475.0271.7971.8874.8772.8267.95
3800071.2574.8371.9866.9574.8076.8375.5372.0371.6974.0572.3768.63
4000072.8876.4674.0968.0875.7577.5376.1373.5872.7075.5573.7068.83
4200073.2376.3074.6868.7175.9977.4176.9673.5971.9675.3473.2167.33
4400073.1376.1974.4968.7074.8677.6573.2673.6771.5774.8372.5967.28
4600072.8276.0074.9367.5374.4676.1974.3372.8571.8774.5873.4167.61
4800073.0176.4674.0768.4974.6276.1674.4173.3071.5974.1973.3967.20
5000072.8476.3574.6967.4874.6476.4073.2674.2571.4474.3272.3267.69

The detection results obtained by the methods proposed in this paper are shown in Table 1.