The plant detection methods were studied by field pictures of rice and image processing. Combined improved normalized difference index of blue-green IDNBG and chromatic model, extracted of plant leaves were extracted by classification, recognition and image threshold segmentation. Forward and reverse morphological operation was used to repair integrity of inside blade. 4-connectivity chain code was used to detect leaf edge and repair smoothly. 40 pictures were taken under the conditions of visible and field. The average correct extraction rate of plant leaves was 83.07%, and the average error extraction rate 3.57%. 90 edge lines were repaired smoothly which resulted in the reduction of correct extraction rate of leaves by 0.63%. The main factors of plant detection were analyzed. It showed that imaging different conditions affected the brightness factor which served as main factor of chromatic model. Morphological dilation operator and forward and reverse filter operator remained some small dew and lesion, so that inside of extracted blade was complete. The chain code could smooth the blade edges and remove some part of correct blades.