Machine Learning–Based Detection and Sorting of Multiple Vegetables and Fruits

Machine Learning Application in Horticulture and Prospects for Predicting Fresh Produce Losses and Waste: A Review

The current review examines the state of knowledge and research on machine learning (ML) applications in horticultural production and the potential for predicting fresh produce losses and waste. Recently, ML has been increasingly applied in horticulture for efficient and accurate operations. Given the health benefits of fresh produce and the need for food and nutrition security, efficient horticultural production and postharvest management are important. This review aims to assess the application of ML in preharvest and postharvest horticulture and the potential of ML in reducing postharvest losses and waste by predicting their magnitude, which is crucial for management practices and policymaking in loss and waste reduction. The review starts by assessing the application of ML in preharvest horticulture. It then presents the application of ML in postharvest handling and processing, and lastly, the prospects for its application in postharvest loss and waste quantification. The findings revealed that several ML algorithms perform satisfactorily in classification and prediction tasks. Based on that, there is a need to further investigate the suitability of more models or a combination of models with a higher potential for classification and prediction. Overall, the review suggested possible future directions for research related to the application of ML in postharvest losses and waste quantification.

A novel method for vegetable and fruit classification based on using diffusion maps and machine learning

Vegetable and fruit classification can help all links of agricultural product circulation to better carry out inventory management, logistics planning and supply chain coordination, and improve the efficiency and response speed of the supply chain. However, the current classification of vegetables and fruits mainly relies on manual classification, which inevitably introduces the influence of human subjective factors, resulting in errors and misjudgments in the classification of vegetables and fruits. In response to this serious problem, this research proposes an efficient and reproducible novel model to classify multiple vegetables and fruits using handcrafted features. In the proposed model, preprocessing operations such as Gaussian filtering, grayscale and binarization are performed on the pictures of vegetables and fruits to improve the quality of the pictures; statistical texture features representing vegetable and fruit categories, wavelet transform features and shape features are extracted from the preprocessed images; the feature dimension reduction method of diffusion maps is used to reduce the redundant information of the combined features composed of statistical texture features, wavelet transform features and shape features; five effective machine learning methods were used to classify the types of vegetables and fruits. In this research, the proposed method was rigorously verified experimentally and the results show that the SVM classifier achieves 96.25% classification accuracy of vegetables and fruits, which proves that the proposed method is helpful to improve the quality and management level of vegetables and fruits, and provide strong support for agricultural production and supply chain.

JujubeNet: A high-precision lightweight jujube surface defect classification network with an attention mechanism

Surface Defect Detection (SDD) is a significant research content in Industry 4.0 field. In the real complex industrial environment, SDD is often faced with many challenges, such as small difference between defect imaging and background, low contrast, large variation of defect scale and diverse types, and large amount of noise in defect images. Jujubes are naturally growing plants, and the appearance of the same type of surface defect can vary greatly, so it is more difficult than industrial products produced according to the prescribed process. In this paper, a ConvNeXt-based high-precision lightweight classification network JujubeNet is presented to address the practical needs of Jujube Surface Defect (JSD) classification. In the proposed method, a Multi-branching module using Depthwise separable Convolution (MDC) is designed to extract more feature information through multi-branching and substantially reduces the number of parameters in the model by using depthwise separable convolutions. What's more, in our proposed method, the Convolutional Block Attention Module (CBAM) is introduced to make the model concentrate on different classes of JSD features. The proposed JujubeNet is compared with other mainstream networks in the actual production environment. The experimental results show that the proposed JujubeNet can achieve 99.1% classification accuracy, which is significantly better than the current mainstream classification models. The FLOPS and parameters are only 30.7% and 30.6% of ConvNeXt-Tiny respectively, indicating that the model can quickly and effectively classify JSD and is of great practical value.

Intelligent grading method for walnut kernels based on deep learning and physiological indicators

Walnut grading is an important step before the product enters the market. However, traditional walnut grading primarily relies on manual assessment of physiological features, which is difficult to implement efficiently. Furthermore, walnut kernel grading is, at present, relatively unsophisticated. Therefore, this study proposes a novel deep-learning model based on a spatial attention mechanism and SE-network structure to grade walnut kernels using machine vision to ensure accuracy and improve assessment efficiency. In this experiment, we found through the literature that both the lightness (L* value) and malondialdehyde (MDA) contens of walnut kernels were correlated with the oxidation phenomenon in walnuts. Subsequently, we clustered four partitionings using the L* values. We then used the MDA values to verify the rationality of these partitionings. Finally, four network models were used for comparison and training: VGG19, EfficientNetB7, ResNet152V2, and spatial attention and spatial enhancement network combined with ResNet152V2 (ResNet152V2-SA-SE). We found that the ResNet152V2-SA-SE model exhibited the best performance, with a maximum test set accuracy of 92.2%. The test set accuracy was improved by 6.2, 63.2, and 74.1% compared with that of ResNet152V2, EfficientNetB7, and VGG19, respectively. Our testing demonstrated that combining spatial attention and spatial enhancement methods improved the recognition of target locations and intrinsic information, while decreasing the attention given to non-target regions. Experiments have demonstrated that combining spatial attention mechanisms with SE networks increases focus on recognizing target locations and intrinsic information, while decreasing focus on non-target regions. Finally, by comparing different learning rates, regularization methods, and batch sizes of the model, we found that the training performance of the model was optimal with a learning rate of 0.001, a batch size of 128, and no regularization methods. In conclusion, this study demonstrated that the ResNet152V2-SA-SE network model was effective in the detection and evaluation of the walnut kernels.

Grading detection of “Red Fuji” apple in Luochuan based on machine vision and near-infrared spectroscopy

A quality detection system for the “Red Fuji” apple in Luochuan was designed for automatic grading. According to the Chinese national standard, the grading principles of apple appearance quality and Brix detection were determined. Based on machine vision and image processing, the classifier models of apple defect, contour, and size were constructed. And then, the grading thresholds were set to detect the defective pixel ratio t, aspect ratio λ, and the cross-sectional diameter Wp in the image of the apple. Spectral information of apples in the wavelength range of 400 nm~1000 nm was collected and the multiple scattering correction (MSC) and standard normal variable (SNV) transformation methods were used to preprocess spectral reflectance data. The competitive adaptive reweighted sampling (CARS) algorithm and the successive projections algorithm (SPA) were used to extract characteristic wavelength points containing Brix information, and the CARS-PLS (partial least squares) algorithm was used to establish a Brix prediction model. Apple defect, contour, size, and Brix were combined as grading indicators. The apple quality online grading detection platform was built, and apple’s comprehensive grading detection algorithm and upper computer software were designed. The experiments showed that the average accuracy of apple defect, contour, and size grading detection was 96.67%, 95.00%, and 94.67% respectively, and the correlation coefficient Rp of the Brix prediction set was 0.9469. The total accuracy of apple defect, contour, size, and Brix grading was 96.67%, indicating that the detection system designed in this paper is feasible to classify “Red Fuji” apple in Luochuan.