Dataset DOI: 10.5061/dryad.cvdncjtfm

Description of the data and file structure

The main goal of the study was to develop a practical and low-cost yield monitoring system for manually harvested strawberries to encourage widespread adoption. A key decision in the system design was to utilize an affordable GPS that relies on the freely and widely available Satellite Based Augmentation System (SBAS) corrections. The pickers were not trained prior to using the iCarritos and were deployed in a commercial field with minimal disruption to established harvesting practices.

Data was collected as part of the USDA Agricultural Research Service (USDA-ARS) Areawide Pest Management Grant Program project "Site-Specific Soil Pest Management in Strawberry and Vegetable Cropping Systems Using Crop Rotation and a Needs-Based Variable Rate Fumigation Strategy” through Non-Assistance Cooperative Agreements.

The publicly released dataset consists of two main folders, SantaMaria_CNNLSTM and SantaMaria_2024, along with corresponding Excel files that record harvested trays. The SantaMaria_CNNLSTM folder was used to train and evaluate the CNN–LSTM model, as well as for picker efficiency and tray fill time computation and evaluation. The SantaMaria_2024 folder was used to compute picker efficiency and tray fill time over the entire harvest season. The following section provides a detailed description of each component.

1. SantaMaria_CNNLSTM.zip:
   The cart_data subfolder within this directory contains the annotated harvest data used for CNN–LSTM model training and evaluation. It includes multiple CSV files, each corresponding to a specific harvest day, following the naming convention mm-dd-yy_train-ready_all_carts.csv (e.g., 4-10-24_train-ready_all_carts.csv). Each CSV file contains the following columns:

   • date_cartID: Combined harvest date and iCarrito identifier.
   • GPS_TOW: GPS Time of Week (milliseconds).
   • easting, northing: UTM coordinates (metres).
   • ax, ay, az: Acceleration along the x, y, and z axes (m s⁻²).
   • raw_mass: Raw harvest mass (kg).
   • activity: Annotated activity label, either “Pick” or “NoPick”.

   The break_log subfolder within this directory contains daily break information for each iCarrito. Files follow the naming format mm-dd-yy_break_log.csv and include:

   • harvest_date: Harvest date.
   • cart_id: iCarrito identifier.
   • no_breaks: Number of breaks taken by the corresponding iCarrito.

   The accompanying Excel file, harvested_trays.xlsx, records the number of trays harvested per iCarrito for each harvest day, with following columns:

   • cart_id: iCarrito identifier.
   • no_trays: Number of trays harvested by that iCarrito.

2. SantaMaria_2024.zip:
   This folder contains the complete season-long harvest data from all iCarritos, including break logs and harvested tray records. The subfolders are organised by date (e.g., DD-MM-YY/) and contain raw harvest (cart1.csv, cart2.csv, …, cartN.csv) corresponding to individual carts. Each CSV file contains ten columns:

   • rpi_utc_time: Raspberry Pi UTC timestamp.
   • gps_utc_time: GPS UTC timestamp.
   • GPS_TOW: GPS Time of Week (milliseconds).
   • LAT, LON, HEIGHT: Geographic coordinates (degrees and metres).
   • ax, ay, az: Acceleration along the x, y, and z axes (m s⁻²).
   • raw_mass: Raw harvest mass (kg).

   The break_log subfolder and the Excel file harvested_trays_seasonlong.xlsx follow the same structure and naming conventions as described above, containing daily break records and total harvested trays for all carts across the season.

The dataset provides comprehensive, annotated time-series data suitable for training and evaluating activity recognition models, as well as for further research in harvest process optimisation and labour efficiency analysis.

Dataset directory structure

datasets/
├── SantaMaria_CNNLSTM/
│   ├── cart_data/
│   │   ├── MM-DD-YY_train-ready_all_carts.csv
│   │   ├── ...
│   │   └── MM-DD-YY_train-ready_all_carts.csv
│   ├── break_log/
│   │   ├── MM-DD-YY_break_log.csv
│   │   ├── ...
│   │   └── MM-DD-YY_break_log.csv
│   └── harvested_trays.xlsx
│
├── SantaMaria_2024/
│   ├── DD-MM-YY/
│   │   ├── cart1.csv
│   │   ├── cart2.csv
│   │   ├── ...
│   │   └── cartN.csv
│   ├── break_log/
│   │   ├── MM-DD-YY_break_log.csv
│   │   ├── ...
│   │   └── MM-DD-YY_break_log.csv
│   └── harvested_trays_seasonlong.xlsx

Code/software

https://github.com/uddhavbhattarai/HumanActivityRecognitionEfficiencyEstimation

Human subjects data

While this dataset is derived from field operations involving human workers, it does not contain any personally identifiable information (PII). The dataset consists solely of sensor data collected from instrumented strawberry-picking carts (iCarritos), including GPS coordinates, timestamps, IMU measurements, and load-cell readings. No personal attributes, names, worker identifiers, or demographic information were collected. Carts were used interchangeably by multiple pickers during commercial harvesting operations, and no individual was permanently or uniquely associated with any specific cart. As a result, there is no linkage between the recorded data and any identifiable person.