Fujio Tsutsumi

Automatic corrosion estimation using galvanized steel images on power transmission towers

We propose a method that automatically estimates the degree of corrosion of galvanized steel of power transmission towers using digital images of the steel. Electric power companies have to determine the corrosion degree of steel for the maintenance of the towers. Accordingly, the technique to estimate the degradation degree of galvanization objectively and nondestructively is need. Our method is based on a support vector machine using the radial basis function kernel. Moreover, we define a new technique that adds artificially processed images to the training data, to handle the variations in color using automatic white balance and exposure of a digital camera. We evaluated our method using 1,427 images of 8 towers. Our method achieved an average precision of 85.6% using a hue-saturation-value histogram feature vector.

A Method to Recognize and Count Leaves on the Surface of a River Using User's Knowledge about Color of Leaves

This paper proposes a recognition and count method of leaves on the surface of a river to be used in mangrove ecosystem monitoring. Conventionally counting leaves required considerable manual labor for precise monitoring of material flow in the ecosystem. Therefore an efficient counting method was needed. Our method automatically recognizes and counts the number of floating leaves in recorded video using color and motion features. The color feature is represented by 3 dimensional histogram of a color space. We have developed a user interface based on the interactive machine learning model to acquire the color feature from video images. The user can easily produce a huge number of sample data to extract the color feature by the user interface in the same way as coloring a picture. For the motion feature, speed and acceleration of the targets are used. The counting method proposed in this paper has been applied to three videos (total five hours) which recorded about 20,000 leaves, and high recall and precision rates of 96% and 94%, respectively, have been achieved.

Distortion-Based Visualization for Long-Term Continuous Acoustic Monitoring

Visualizing long-term acoustic data has been an important subject in the field of equipment surveillance and equipment diagnosis. This paper proposes a distortion-based visualization method of long-term acoustic data. We applied the method to 1 hour observation data of electric discharge sound, and our method could visualize the sound data more intelligibly as compared with conventional methods.

Sound Visualization and Retrieval Technique for Assisting Hearing Memory of Patrol Worker

In this paper, we propose a system to support equipment check patrol. One of the most difficult tasks for patrol worker is discriminating the difference of equipment sound in the check spot. The main purpose of our system is to support human hearing by automatic segmentation, visualization and retrieval of equipment sound. Automatic segmentation function is realized using a new video shot segmentation method. The method uses fluctuating thresholds according to time transition of visual changes, and performs automatic segmentation of recorded sound according to the step of check work. The visualization and the retrieval functions are based on the spectrum subtraction method and analyzing technology of the temporal frequency map. This paper shows the evaluation result of the segmentation method using equipment check video. And it also shows the effectiveness of the proposed visualizing and retrieval method through an example of electric discharge sound.