Search

Challenges exist in the power generation efficiency and safety of marine current turbines (MCTs), as the MCT blades are often attached by foreign objects when operating underwater. It is essential for the stable operation of an MCT to recognize attachments timely and accurately. However, underwater imaging suffers from blurry edges due to light attenuation and scattering. It is challenging for accurate recognition through underwater images since blurry edges result in unclear edge features. To alleviate this problem, LAW-IFF Net is proposed in this paper, which mainly contains two parts. Firstly, this paper proposes to transform the local averages of feature maps into weight matrices, namely the locally average weighting (LAW) mechanism. It is intended to alleviate the edge gradient reduction caused by blurry edges. Secondly, the proposed improved feature fusion (IFF) mechanism aims to overcome the deviation caused by the feature fusion of different attention branches based on spatial attention. At the same time, the lightweight networks are combined with the proposed method to improve the computation speed and ensure the timeliness of recognition. Experimental results on the MCT dataset show the superiority of the proposed method in terms of accuracy and speed of attachment recognition in images with blurry edges. The experimental results on publicly available datasets show the applicability of the proposed method to other underwater images. [ABSTRACT FROM AUTHOR]

To diagnose the attachment of marine current turbine, this article proposes a method based on convolutional neural network and the concepts of depthwise separable convolution to achieve feature extraction. The method consists of three steps: data preprocessing, feature extraction and fault diagnosis. This method can diagnose the fault degree of blade imbalance and uniform attachment in underwater environment with strong currents and complex spatiotemporal variability. It can extract distinct image feature in harsh marine environments by using a convolutional neural network. In addition, this method is robust for the recognition of blurred pictures under high-speed rotation. [ABSTRACT FROM AUTHOR]

According to the mechanism of rice growth, if nitrogen deficiency occurs, not only rice leaf but also sheath shows special symptoms: sheaths become short, stems appear light green, older sheath become lemon-yellowish. Nitrogen nutrition status of rice could be identified based on the differences of color and shape of leaf and sheath under different levels of nitrogen nutrition. Machine vision technology can be used to non-destructively and rapidly identify rice nutrition status, but image acquisition via digital camera is susceptible to external conditions, and the images are of poor quality. In this research, static scanning technology was used to collect images of rice leaf and sheath. From those images, 14 color and shape characteristic parameters of leaf and sheath were extracted by R, G, B mean value function and region props function in MATLAB. Based on the relationship between nitrogen content and the characteristics extracted from the images, the leaf R, leaf length, leaf area, leaf tip R, sheath G, and sheath length were chosen to identify nitrogen status of rice by using Support Vector Machine (SVM). The results showed that the overall identification accuracies of different nitrogen nutrition were 94%, 98%, 96% and 100% for the four growth stages, respectively. Different years of data were used for validation, identification accuracies were 88%, 98%, 90% and 100%, respectively. The results showed that additional sheath characteristics can effectively increase the identification accuracy of nitrogen nutrition status and the methodology developed in the study is capable of identifying nitrogen deficiency accurately in the rice. [ABSTRACT FROM AUTHOR]