Fruit Sizing in Orchard: A Review from Caliper to Machine Vision with Deep Learning

Abstract

Forward estimates of harvest load require information on fruit size as well as number. The task of sizing fruit and vegetables has been automated in the packhouse, progressing from mechanical methods to machine vision over the last three decades. This shift is now occurring for size assessment of fruit on trees, i.e., in the orchard. This review focuses on: (i) allometric relationships between fruit weight and lineal dimensions; (ii) measurement of fruit lineal dimensions with traditional tools; (iii) measurement of fruit lineal dimensions with machine vision, with attention to the issues of depth measurement and recognition of occluded fruit; (iv) sampling strategies; and (v) forward prediction of fruit size (at harvest). Commercially available capability for in-orchard fruit sizing is summarized, and further developments of in-orchard fruit sizing by machine vision are anticipated.

Management Information Systems for Tree Fruit—1: A Review

“…The motivation was provided by the recent progress in systems for the in-orchard assessment of tree-fruit numbers and the size profile based on machine vision and for a forecast of optimum harvest dates from enhanced heat unit calculations and estimates of dry matter content of fruit-on-tree using NIR spectroscopy-as reviewed by Anderson et al [18], Neupane et al [21], Amaral et al [22] and Walsh et al [23], respectively. In our own experience in developing these technologies and prompting their use by growers, it became obvious that the output of the new technologies needed to be delivered into a MIS to be used by the orchard manager.…”

“…For example, a deep learning model for an Android smartphone app, "KiwiDetector", was developed for the yield estimation of kiwifruit [101], a smartphone camera application was used in a client-server architecture for the estimation of apple yields [90], and the orange fruit count was estimated from UAV collected imagery, with results broadcast to an online map [102]. The use of depth cameras or LiDAR has enabled the estimation of size profiles of fruit on trees, with a coupling to a fruit growth model that allows a forecast of the fruit size distribution at harvest, as reviewed by Neupane et al [21]. Moving beyond the provision of yield data per se to use in management, a system for estimation of the optimal number of harvest containers and their field placement for efficient logistics was developed based on apple yield mapping from orchard video images [56].…”

Management Information Systems for Tree Fruit–2: Design of a Mango Harvest Forecast Engine

“…Other approaches for harvest-load estimation use satellite-imagery-derived vegetation indices [15,16] and UAV-derived canopy structure attributes [8]. Our research group has reviewed each of these aspects, i.e., the forecast of harvest timing based on GDD [17] and/or DMC [18], the forecast of fruit number [3,19], and the forecast of fruit size at harvest [20].…”

“…There is an allometric relationship between mango fruit lineal dimensions and fruit mass [20]. Fruit lineal dimensions can be collected manually, using calipers, and a statistically robust sampling regime, with data capture aided with a mobile device.…”

Fast region based convolutional neural network ResNet-50 model for on tree Mango fruit yield estimation

“…Some of the significant issues encountered throughout the study were prominent flushed regions, occlusions from leaves, stems, and branches, as well as non-uniform lighting in distinct fruit segments [15]. Study proposes a review of over the past thirty years, of automating the process of sizing fruits and vegetables by moving from mechanical means to machine vision [16]. Proposed work suggests a visual object tracking network called YOLO-deepsort to count and identify tomatoes at various stages of growth.…”