Abstract
Crop-load or yield estimation is an important management tool for growers of fresh market fruit crops such as apples. Estimation of expected harvest volume is crucial for efficient management of pre- and post-harvest operations such as acquiring desired equipment and labor force for harvesting and transporting fruit from orchards to packing houses. There are no commercial equipment or platform that can automatically estimate the total number of fruit and fruit size in orchards. Washington State, which is the top producer and supplier of apples in the United States, produces 15-18 billion apples every year. Estimation of such vast quantities of fruit is often done using crop forecasting models, which is time-consuming, as different parameters have to be measured and estimated for individual orchards. The performance of this method is limited by variability in climate, cultivar, and geographic location among others.
Current research focuses on automated counting and sizing of fruits using camera images of tree canopies and machine vision algorithms. Often, these computations are accomplished off-line after images are acquired. Machine vision systems currently being used in research are heavy and cumbersome, and require specific skills to operate, which can be expensive for growers for commercial adoption. In the past several years, we have worked on machine vision system to detect, and localize apples for robotic harvesting in orchard environment. The machine vision algorithm developed during this work has also been used for crop-load estimation with good results. Our new approach, being disclosed here, is to adopt our existing vision algorithm on to a mobile or smartphone platform. The designed smartphone application will be used to acquire apple tree canopy images at desired sample locations in orchards (based on principles of spatial statistics), and detect, count and size fruit to estimate crop-load in orchards in near real-time.
Applications and Advantages
Because smartphones are ubiquitous and pre-equipped with all necessary sensors such as camera(s) and GPS, an app-based approach has great potential for on-hand, near-real time, and low cost crop-load estimation for apple or tree fruit growers.