Mobarakeh steel company is the biggest steel maker company in Iran. Iection of different units of this company (e.g. pelletizing plant) is done by human, such that the quality is enhanced and the probabilistic costs are decreased. In the peletizing unit, pellets are posed on the pallets with 150 cm width and 360 cm length. The pellets undergo drying and preheating in the stove. There are four rows of grate bars on the surface of the pallet. Each row consists of 90 grate bars. Gradually the grate bars will be damaged because of the high temperature of stove, sudden change of temperature and also because the pellets hit the grate bars. The damages cause spaces between grate bars and this lead to losses in pelletizing unit. If we observe and report the damages periodically we can prevent these losses through replacing the pallets if required. For this reason pallets need permanent iection. Today this iection is done by the human. In this thesis, a method for automatic detection of these damages is presented. At first, the pallet area is segmented from the image by three proposed algorithms and the pallet is divided to four areas such that each area has a row of grate bars. Then in the image of one row of the grate bars the space between every two adjacent grate bars is segmented that we call them objects. Therefore, we use four methods: segmentation according to statistical information of the color of images, segmentation using k-means algorithm, segmentation using local k-means algorithm and a proposed multistage segmentation algorithm. Then 103 features (color and shape feature) are extracted from the objects. Among these features 33 more effective features are selected using logistic regression. Finally, based on these selected features the objects are Key words Pallet, grate bars, segmentation, feature extraction, feature selection, dir=rtl
