In this paper, an overview of the challenges for the Unmanned Aerial Vehicles path planning for the inspection of olive orchards will be given. Standard approaches, limitations and requirements for specific types of applications such as olive fruit detection and yield prediction are analyzed. Possible flight planning approaches are compared along with their advantages and disadvantages. An optimal solution is proposed and expressions for estimating the recorded canopy area of an individual tree are derived. The values calculated in this way using the described fruit detection method allow for the estimation of the total yield.

Accurate fruit yield estimation is increasingly recognized as a vital component of sustainable agricultural practices, particularly in economically and culturally significant crops like olives. As the global demand for olive oil rises, so does the need for precise yield forecasting, which can lead to better resource allocation and enhanced management practices. The advent of machine learning (ML), particularly deep neural networks (DNNs) and computer vision, has revolutionized agricultural yield estimation practices. 
Thus, in this paper, we explore several state-of-the-art DNN architectures and their performance on two databases: our custom one (with 11,227 olive instances) and a database available from the literature (with 245,089 olive instances). Obtained results (mAP50 ranging from 68% to 95% depending on the used architecture) demonstrate the applicability of the DNN-based approaches in the complex and difficult task of olive detection from high-resolution images, while highlighting the importance of database size and quality (diversity) on DNN's final performance. 

Dataset: https://zenodo.org/search?q=Split_Olive_DATASETv1&l=list&p=1&s=10&sort=bestmatch

Accurate detection and localization of olives in orchard environments is essential for precision agriculture, enabling tasks such as yield estimation, automated harvesting, and crop monitoring. While YOLOv8 provides robust object detection capabilities, optimizing its performance for olive detection remains challenging due to small object sizes, visual variability, and labeling costs. This paper explores the benefits of integrating active learning with YOLOv8, focusing on uncertainty-based sampling and novel missed-detection strategies. We propose and evaluate three acquisition methods—missed ground-truth detection, low-confidence sampling and a small-size (fuzzy score) sampling - to minimize annotation effort while maximizing detection accuracy. Experimental results show that careful hyperparameter tuning, combined with a strategic active learning pipeline, can improve mean Average Precision (mAP) by 7–10 percentage points and reduce labeling effort by 25–35%. Our findings provide practical insights for deploying efficient olive detection systems in real-world agricultural contexts. 

This study investigates the application of unmanned aerial vehicles with multispectral cameras for monitoring the condition of olive groves with high accuracy. Multispectral images of olive groves provided detailed insight into the spectral data required for the analysis of vegetation indices. Using deep learning-based object detection, individual olive trees were identified within the images, which allowed the extraction of parts corresponding to each tree. To separate the background from the canopy, segmentation based on the monocular depth estimation algorithm, Depth Anything, was applied. In this way, elements that are not part of the tree’s crown were removed for more accurate analysis and calculation of the NDVI (Normalized Difference Vegetation Index) and NDRE (Normalized Difference Red Edge Index) indices. The obtained results were compared with the results obtained for unsegmented patches, threshold-based patches, and manually segmented patches. The comparison and analysis carried out shows that the proposed segmentation approach improved the accuracy of NDVI and NDRE by focusing exclusively on the crowns of the observed trees, excluding the noise of the surrounding vegetation and soil. In addition, measurements were carried out on three observed olive groves at different parts of the vegetation cycle, and the values of the vegetation indices were compared. This integrated method combining drone-based multispectral imaging, deep learning object detection, and advanced segmentation techniques highlights a robust approach to olive tree health monitoring and provides insight into seasonal vegetation dynamics, for winter and spring, to capture differences in vegetative activity. 

Although active learning can reduce the effort required to annotate object detection data, many current methods rely on a single selection criterion or combine criteria without accounting for annotation costs or their interactions. This paper presents a multi-criterion, cost-aware active learning framework for detecting small objects in agricultural images. The framework jointly considers prediction uncertainty, object size, scene density, and annotation cost. We evaluate both scalarized and Pareto-based selection strategies across five cost models and conduct an ablation study to examine the role and interactions of each criterion. Experimental results demonstrate that explicit annotation cost modeling improves active learning efficiency by reducing the amount of annotation required to achieve a given level of detection performance. Across multiple cost formulations and selection strategies, cost-aware acquisition reaches comparable accuracy and reduces the estimated annotation effort required to reach comparable detection performance by up to 50% compared to random sampling, where annotation effort is approximated using prediction-derived cost proxies.

 A practical, reproducible pipeline for olive yield estimation is presented, combining close-range UAV still imagery, automated fruit detection, and a simple harvest-based calibration step. For a chosen set of six test trees, one high-resolution canopy image was captured at multiple dates, a fine-tuned object detector was applied to obtain digital fruit counts, and the median of per-tree detections was used as a robust representative digital count. Harvest weights collected at the end of the season were converted to estimated fruit counts using per-cultivar mean fruit mass, while calibration coefficients were computed as the ratio of harvest-derived counts to the representative digital counts. Calibration coefficients are proposed for two observed olive cultivars, enabling estimation of total per-cultivar yield in the observed orchard. The workflow is evaluated and the primary sources of uncertainty – viewpoint-dependent coverage, occlusion, illumination variation, and detector errors – are analysed. The results demonstrate the feasibility of translating single-view digital counts into calibrated yield estimates, while also highlighting substantial variability that motivates pragmatic mitigations: further in-domain labelling and retraining, multispectral foliage masking, scale-aware architectures, and multi-view sampling. The proposed approach is lightweight and suitable for operational orchard monitoring; recommendations and a roadmap to reduce uncertainty and improve robustness are provided. 

An intelligent system for detection of olive fruits based on implementation of deep neural networks is described. More than 800 images of various olive sorts during several months (from July to November) were collected by UAV. The system will assess the ripeness of individual fruits based on the analysis of selected images.  Also, the system being developed ultimately estimates the expected yield of an individual tree as well as the entire olive grove. Although the annotation process of all collected images is not complete, initial detection results using a significantly smaller number of annotated images (91 images) show good results and provide good input data for the next processing stages. The developed system will ultimately be available to small olive growers as a computer application.