This work introduces a new vision-based approach for estimating chlorophyll contents in a plant leaf using reflectance and transmittance as base parameters. Images of the top and underside of the leaf are captured. To estimate the base parameters (reflectance/transmittance), a novel optical arrangement is proposed. The chlorophyll content is then estimated by using linear regression where the inputs are the reflectance and transmittance of the leaf. Performance of the proposed method for chlorophyll content estimation was compared with a spectrophotometer and a Soil Plant Analysis Development (SPAD) meter. Chlorophyll content estimation was realized for Lactuca sativa L., Azadirachta indica, Canavalia ensiforme, and Lycopersicon esculentum. Experimental results showed that—in terms of accuracy and processing speed—the proposed algorithm outperformed many of the previous vision-based approach methods that have used SPAD as a reference device. On the other hand, the accuracy reached is 91% for crops such as Azadirachta indica, where the chlorophyll value was obtained using the spectrophotometer. Additionally, it was possible to achieve an estimation of the chlorophyll content in the leaf every 200 ms with a low-cost camera and a simple optical arrangement. This non-destructive method increased accuracy in the chlorophyll content estimation by using an optical arrangement that yielded both the reflectance and transmittance information, while the required hardware is cheap.

Plant stress detection involves the process of Identification, Classification, Quantification, and Prediction (ICQP) in crop stress. Numerous approaches exist for plant stress identification; however, a majority rely on expert personnel or invasive techniques. While expert employees demonstrate proficiency across various plants, this approach demands a substantial workforce to ensure the quality of crops. Conversely, invasive techniques entail leaf dismemberment. To overcome these challenges, an alternative is to employ image processing to interpret areas where plant geometry is observable, eliminating the dependency on skilled labor or the need for crop dismemberment. However, this alternative introduces the challenge of accurately interpreting ambiguous image features. Motivated by the latter, we propose a methodology for plant stress detection using 3D reconstruction and deep learning from a single RGB image. For that, our methodology has three steps. First, the plant recognition step provides the segmentation, location, and delimitation of the crop. Second, we propose a leaf detection analysis to classify and locate the boundaries between the different leaves. Finally, we use a Deep Neural Network (DNN) and the 3D reconstruction for plant stress detection. Experimental results are encouraging, showing that our approach has high performance under real-world scenarios. Also, the proposed methodology has 22.86% higher precision, 24.05% higher recall, and 23.45% higher F1-score than the 2D classification method.

Applications such as autonomous navigation, robot vision, and autonomous flying require depth map information of a scene. Depth can be estimated by using a single moving camera (depth from motion). However, the traditional depth from motion algorithms have low processing speeds and high hardware requirements that limit the embedded capabilities. In this work, we propose a hardware architecture for depth from motion that consists of a flow/depth transformation and a new optical flow algorithm. Our optical flow formulation consists in an extension of the stereo matching problem. A pixel-parallel/window-parallel approach where a correlation function based on the sum of absolute difference (SAD) computes the optical flow is proposed. Further, in order to improve the SAD, the curl of the intensity gradient as a preprocessing step is proposed. Experimental results demonstrated that it is possible to reach higher accuracy (90% of accuracy) compared with previous Field Programmable Gate Array (FPGA)-based optical flow algorithms. For the depth estimation, our algorithm delivers dense maps with motion and depth information on all image pixels, with a processing speed up to 128 times faster than that of previous work, making it possible to achieve high performance in the context of embedded applications.