Limited receptive area neural classifier based image recognition in micromechanics and agriculture

Two multi-purpose image recognition systems based on neural classification are proposed. First application is an image recognition based adaptive control system for micromechanics where the limited receptive area (LIRA) neural classifier is proposed for texture recognition of mechanically treated metal surfaces. The performance of the proposed classifier was tested on image database of 80 images of four texture types corresponding to metal surfaces after milling, polishing with sandpaper, turning with lathe and polishing with file. The promising recognition rate of 99.8% was obtained for image database divided in half into training and validation sets. The second application is agriculture, where vast amounts of pesticides are used against the insects. In order to decrease the required amount of pesticides it is necessary to locate the precise form distribution of the insects and larvae. In this case the use of pesticides will be local. In order to automate the task of recognition of larvae we propose to use a web-camera based computer vision system. We tested the proposed system on image database of larvae of different forms, sizes and colors, distributed in different amounts and positions and containing 79 images. The main idea was to recognize the difference between the textures corresponding to the larvae and real world background. Recognition rate of 90.36% was obtained with only 10 images used for training and the rest of the image database used for validation suggesting high potential of the proposed approach. In this article we present the structure and the algorithms of recognition with LIRA neural classifier, and results of its application in both tasks. Keywords—agriculture, image recognition, larvae, LIRA, micromechanics, neural classifier.