Research On The Intelligent Image Generation And Classification For Ship Coating Defects Method

As one of the three pillars of modern ship building process,ship coating runs through the whole shipbuilding process from design and construction until the ship is delivered.The quality of the coating is directly related to the construction cycle and maintenance cost of the ship as well as an important factor affecting the anti-corrosion performance of the hull and the service life of the ship.At present,the transformation and upgrading of the shipbuilding industry driven by intelligent manufacturing is in its initial stage.Aiming at the lack of scientific and reasonable intelligent detection method of coating defects in most domestic shipyards at this stage,small number and class imbalance of coating defect images,research on the intelligent image generation and classification method for ship coating defect was conducted.On this basis,research on zero-shot classification was carried out to realize the intelligent generation and classification of ship coating defects,so as to reduce workload,this improving the intelligent level of the domestic shipbuilding industry.The main research work of the thesis is as follows:(1)As the sources of coating defects in each category collected at present were different,the number of defect categories varied greatly,which in turn reduced the detection performance of intelligent detection equipment,a novel intelligent generation method of unbalanced ship coating defect images based on IGASEN-EMWGAN was proposed for generating new images of a minority classes of ship coating defects to obtain balanced datasets of ship painting defect.The experimental results showed that the IS value and FID score value were improved and reduced by 4.92% and 7.29%,respectively,compared with the baseline generation model.Based on the experimental results,the intelligent generation of ship defect images was achieved.(2)Because of the high labelling cost,the current number of ship coating defects was insufficient,which in turn reduced the detection performance of intelligent detection equipment,a novel intelligent classification method of small-sample ship painting defects images based on DCCVAE-IACWGAN was proposed for detecting the known categories of ship coating defects.The experimental results showed that the proposed model had obtained92.54% accuracy,88.33% F1-score,and 91.93% G-mean.Based on the experimental results,the classification of known classes of ship coating defects was achieved.(3)Due to the complexity and uncertainty of ship coating conditions,the new unknown defects without labels were recognized as known defects by intelligent inspection equipment,thus reducing its detection performance,a zero-shot classification method based IDATLWGAN was proposed for detecting the emerging unknown categories of ship coating defects.The experimental results showed that the overall performance of the model is better than other existing transfer learning models.Based on the experimental results,the classification of emerging unknown classes of ship coating defects was achieved.