Estimating harvests with image-based fruit detection and intelligent data analysis

Maximizing the proportion of top-quality fruits is a major challenge in fruit production, where 20-40% of fruits are downgraded due to size and color issues. One key factor influencing fruit quality is the fruit load, which producers manage early in the season using chemical thinning agents. However, as their effectiveness is influenced by weather conditions, additional manual thinning, although laborious, is necessary in most cases. Therefore, developing an accurate crop yield estimation tool is crucial for optimizing resource use and increasing profits. Currently, in the absence of automated systems, producers rely on time-consuming manual counting, leading to potential errors of up to 66%. In addition to management strategies, final yield is also affected by climate conditions and plant growth, which are not always considered in current estimation methods, making accurate yield predictions difficult.

This project aims to create a simple and effective tool for highly accurate yield estimation for fruit growers based on smartphone images of apple trees combined with fruit diameter measurements and micro-climatic data. The tool will be integrated into the IoT platform developed by HEIG-VD as part of the project Novel. To reach the project objective this innovative approach will combine image analysis and machine learning techniques to automatically estimate the number of fruits on the one hand and a fruit growth model built upon the direct measurement of fruit size using connected dendrometers on the other. The existing IoT platform also continuously collects climate and soil moisture data, which will be integrated into the model for more accurate yield predictions. The mobile app, incorporating the capture of images, will be designed to be user-friendly, ensuring that farmers can easily adopt and use it in their daily routines.