Estimation of crop kinds and growth stages in vegetable fields using high resolution satellite images

Abstract The estimation of area size of vegetable fields which are covered by different crop kinds and growth stages, is indispensable to keep supply and demand in balance for Japanese agricultural market. Few studies have been reported for vegetable crops, while several attempts using satellite images have. been studied for cereal crops, such as rice, wheat, and corn. These cereal crops are cultivated once or twice a year, and each crop is planted simultaneously. Then, the kind and growth stage estimation of cereal crops using satellite images is practicable only with a crop calendar. On the other hand, vegetable crops are planted many times in a year for constant products supply and seasonal labor dispersion, owing to their short growing period. It causes many different growth stages at a same time even for a single kind of vegetation. Therefore, the estimation of both crop kinds and growth stages are necessary because of the unavailability of crop calendars. We have proposed some methods for vegetable crop classification and growth stage estimation using airborne hyperspectral images which have more bands and higher spatial resolution than satellite images, and obtained good results. In this paper, we demonstrate the ability of vegetable crop classification and growth stage estimation using high resolution satellite images that have less spectral information but have better cost performance than airborne hyperspectral images. Our target vegetable is cabbage in three areas, Choshi, Tsumagoi, and Atsumi, where cabbages are produced most for each harvest season in Japan. The accuracy of cabbage identification in luxuriant growth stages, in which the coverage of cabbage is almost 100%, is more than 90% for each area. When all growth stages are included, the accuracy is 76.5% for Choshi, 80.2% for Tsumagoi, and 57.7% for Atsumi. Most errors are caused in fields where the coverage of vegetation is small. Others are caused by misclassification of cabbage as other vegetables, such as purplish cabbage, celery. Besides, Japanese radish, purplish cabbage, cauliflower, and broccoli are misclassified as cabbage. We examined two kinds of growth stage estimation, one is maximum likelihood method and the other is the method using vegetation index. DVVI (Derivative Value for Vegetation Index). The accuracy of growth stage estimation is 53.0% for Choshi, 81.7% for Tsumagoi, and 85.6% for Atsumi. Our method shows usefulness to estimate the yield of cabbage just before the harvest or after the production season, and to monitor the state of growing.