Dataset for worker activity recognition and efficiency estimation during manual harvesting

This dataset contains harvest data collected during manual strawberry harvesting with instrumented picking carts in Santa Maria, CA, USA, in 2024. The data includes geo-tagged harvest mass, cart location, and motion recorded by a GPS receiver, an Inertial Measurement Unit (IMU), and load cells. Each data point is annotated as either "Pick" (indicating active picking) or "NoPick" (indicating no active picking). This dataset can be used to train, validate, and test AI algorithms to recognize worker activity during manual fruit harvesting and quantify worker efficiency. It is valuable for researchers and practitioners in precision agriculture and agricultural automation who are working on optimizing labor and field management, as well as developing strawberry harvesting machines or harvest assist systems.

Methods
Data was collected from instrumented picking carts, iCarrtios, in a commercial strawberry field in Santa Maria, California, growing Fronteras variety. Plants were cultivated in raised beds with a width of 110 cm, each containing four parallel strawberry rows.

iCarritos were developed by instrumenting a wire frame structure picking carts (aka carritos) with a wheelbarrow system. Carts were equipped with a SwiftNav Piksi GNSS unit with an integrated inertial measurement unit (IMU) to record the geospatial position of harvest and cart motion. The GNSS unit had a horizontal Circular Error Probable (CEP) accuracy of 0.75 meters. Two load cells were installed in front and rear of the cart to measure harvest mass. Recorded harvest mass was obtained by averaging the front and rear mass readings.

A Raspberry Pi 0W microcomputer was used as the central processing unit, and an SD card was used to run the carrito software and store the data during the harvest at 10 HZ.