Fig 4 - uploaded by Nikita Stasenko
Content may be subject to copyright.
Source publication
Context in source publication
Context 1
... ground truth, we understand an area that is known to be correctly highlighted, and it is used to check the results of segmentation by given models. In this paper, ground truth segmentation was created manually. Fig. 4, Fig. 5, and Fig. 6 show the examples of original images and their corresponded ground truth (the correct segmentation) of the specified areas in comparison with the results of segmentation by Deeplab and U-Net models, respectively. The examples of segmentation by Mask R-CNN for "Whole apple", "Spoiled areas" and "Cracks" objects are given in Fig. 7: the ...
Similar publications
Fruit texture affects apples’ quality, consumer preference, and shelf life. The cell wall neutral sugar composition was reported to contribute to apples’ mechanical properties at harvest. However, the contributions of cell wall neutral sugar composition to apple texture loss during storage among different cultivars are still unclear. In this study,...
Transportation vibration adversely affects fresh produce during transportation. In addition, storage temperature affects the quality of fresh commodities. The physical changes in Java apples during transportation and storage were evaluated in this study. Java apples were transported from local farms to wholesale markets (180 km). Java apples were s...
Citations
... In this work, we implement a Mask R-CNN model due to the Feature Pyramid Network (FPN) and ResNet101 backbone, which allow for generating the bounding boxes (object detection) and segmentation masks (instance segmentation). In [45], we have compared the Mask R-CNN to such applied CNN-based models as U-Net [46] and Deeplab [47] for early postharvest decay detection, and Mask R-CNN achieved the highest performance in terms of average precision, namely 67.1% against 59.7% and 56.5%, respectively. Moreover, the Mask R-CNN model generates the bounding boxes and segmentation masks of the postharvest decay and fungal zones separately from each other. ...
Food quality control is an important task in the agricultural domain at the postharvest stage for avoiding food losses. The latest achievements in image processing with deep learning (DL) and computer vision (CV) approaches provide a number of effective tools based on the image colorization and image-to-image translation for plant quality control at the postharvest stage. In this article, we propose the approach based on Generative Adversarial Network (GAN) and Convolutional Neural Network (CNN) techniques to use synthesized and segmented VNIR imaging data for early postharvest decay and fungal zone predictions as well as the quality assessment of stored apples. The Pix2PixHD model achieved higher results in terms of VNIR images translation from RGB (SSIM = 0.972). Mask R-CNN model was selected as a CNN technique for VNIR images segmentation and achieved 58.861 for postharvest decay zones, 40.968 for fungal zones and 94.800 for both the decayed and fungal zones detection and prediction in stored apples in terms of F1-score metric. In order to verify the effectiveness of this approach, a unique paired dataset containing 1305 RGB and VNIR images of apples of four varieties was obtained. It is further utilized for a GAN model selection. Additionally, we acquired 1029 VNIR images of apples for training and testing a CNN model. We conducted validation on an embedded system equipped with a graphical processing unit. Using Pix2PixHD, 100 VNIR images from RGB images were generated at a rate of 17 frames per second (FPS). Subsequently, these images were segmented using Mask R-CNN at a rate of 0.42 FPS. The achieved results are promising for enhancing the food study and control during the postharvest stage.
... Thanks to this progress, machine learning and deep learning methods have been used frequently in many fields, such as defense industry, meteorology, psychology, medicine, industrial applications, energy systems, and agriculture. With the help of machine learning and deep learning-based applications, farmers can keep the crop production process under control, both in the natural environment and in greenhouse environments [10,11]. In general, machine learning applications are actively used in many different areas, such as plant quality diagnosis, soil analysis, insect diagnosis, disease detection, and treatment to improve agriculture [12]. ...
Agricultural product quality assessment is crucial for determining marketability and managing waste. It helps ensure that products meet industry standards and consumer expectations, leading to increased sales and reduced spoilage. It is important for growers, processors, and distributors to have systems in place to ensure the quality of their agricultural products. When the literature is examined, it is seen that artificial intelligence is used for classification processes of many agricultural products. Many studies carried out in this context are based on processing images of agricultural products with various deep learning and machine learning methods and classifying their quality according to these results. In this study, data sets consisting of statistical properties obtained by GLCM, Color Space, and Morphological methods were combined for the first time in this study and used as a single data set. In addition, hybrid classification processes were carried out by applying dimension reduction methods, such as Stacked AutoEncoder, ReliefF, and deep learning methods, such as CNN, SVC, Ridge Classifier, and Subspace Discriminant, to the created data set. When morphological features were given as input to ML algorithms for normal classification, the SAE–CNN hybrid model we proposed in the study achieved a success above the literature with 98.96% accuracy using 32 features. The experimental results demonstrated the effectiveness of the proposed lemon classification system.
... weed recognition [41], pest counting and identification [42], harvest time prediction for tomato plants [43], or apple fruit detection in-field [44]. In [45] the authors reported that Mask R-CNN showed the best evaluation performance at the validation stage in terms of Average Precision (AP) metric comparing to U-Net and Deeplab, namely 67.1% against 59.7% and 56.5%, respectively, for the detection of postharvest decay zones in apples using the sequential RGB imaging data. ...
... Table IV shows the AP values performed by Mask R-CNN for selected images subsets during the validation stages. In comparison with the results shown in [45], we achieved higher evaluation results for Mask R-CNN for postharvest areas detection and segmentation in apples in mAP metrics (67.100 and 87.775, respectively.) In this work, we trained the model separately on each images subset, while in [45] the model was trained on 4440 images. ...
... In comparison with the results shown in [45], we achieved higher evaluation results for Mask R-CNN for postharvest areas detection and segmentation in apples in mAP metrics (67.100 and 87.775, respectively.) In this work, we trained the model separately on each images subset, while in [45] the model was trained on 4440 images. Next, we tested trained weights on DMD reconstructed images to get prediction masks for postharvest decay areas in apples. ...
The damages and diseases that may occur in apples during the storage and cannot be seen visually at early stages is a significant problem in precision agriculture leading to the loss of the lion share of crop. The difficulty of predicting postharvest degradation of apples while in storage is addressed in this article. For the prediction of postharvest decay zones, we used the Dynamic Mode Decomposition (DMD) method in conjunction with the Mask R-CNN model based on Convolutional Neural Networks (CNNs). For validating this idea we have designed a small greenhouse equipped with necessary sensors and actuators for simulating storage conditions for apples. We collected a dataset which includes 552 images of apples stored under the extreme temperatures and humidity. For DMD reconstructed images demonstrating the growth dynamics of decay in apples, the proposed approach achieves 0.999 Structural Similarity Index Measure (SSIM) and 66.813 Peak Signal to Noise Ratio (PSNR). Mask R-CNN predicted postharvest decay zones in apples with an Average Precision (AP) of 83.809% and F1-score with 83.307%. The proposed method is promising for enhancing food storage and postharvest control in precision agriculture since it is based on contactless sensing with video cameras.
... Nowadays, DL based approaches have grown a lot of interest in gaining enhanced accuracy in classifying and predicting healthy and unhealthy fruits along with shelf life. In [Stasenko et al. (2021)], shelf life and decay prediction in apples based on the DL approach is introduced. In [Suzuki et al. (2022)], predicting the shelf life and weight loss in persimmon fruits using the DL approach is emphasized. ...
... Hence, these improved Equations improve the classification performance by considering the fitness function using Equation (16). Figure 3 shows the flowchart of the ISO algorithm. ...
One of the major requirements of agriculture is the quality assessment and ripeness of the agricultural products using non-destructive techniques. The ability of deep learning (DL) models to accurate classification and prediction are increasing. The work’s major aim is to use the DL model to predict the shelf life of pomegranate fruits. Initially, the input MRI pomegranate images are resized to the proper size. Then, the features are extracted and classified using the deep learning model (DL) S-ResNet-152 (Squeeze based ResNet-152). This DL model classifies fruits as healthy or unhealthy. Further, for optimizing the layers and minimizing the loss function, the metaheuristic optimization improved sandpiper optimization (ISO). Then, the healthy fruits are considered for the prediction process. Here, the pomegranate fruit’s shelf life is predicted using the DL model. The features like physiochemical and physiological loss in weight (PLW) and Firmness are predicted for determining the fruit quality. These features are given as input to the hybrid DL model bidirectional gated auto network (Bi-GRU-AN) is used for the prediction of shelf life. The performance of the proposed classification and prediction results are compared with other DL models in terms of the square of correlation coefficient (R²), root mean square error of calibration (RMSEC), and root mean square error of validation (RMSEV).
Food security is one of the priorities of every country in the World. However, different factors are making it difficult to meet global targets on food security. Some unprecedented shocks are encumbering food security at the global level. Various interventions have been applied toward food security and artificial intelligence is one of the modern methods that is being used in various stages of the food system. In this paper, the application of artificial intelligence in the whole food production ecosystem ranging from crop production, livestock production, harvesting/slaughtering, postharvest management, food processing, food distribution, food consumption and food waste management is assessed. The objective of this research is to assess the application of artificial intelligence systems in all the stages of food systems. A systematic review was conducted by analyzing 110 articles after the screening of 450 articles based on the inclusion and exclusion criteria. The results indicated that various artificial intelligence algorithms are being applied to all the stages of the food system from crop/livestock production up to food or agro-waste management.
