Mango

Eight types of fruits and vegetables including lettuce, cabbage, carrot, tomato, green pepper, banana, mango, and salad were purchased from four conveniently selected local markets, namely, “Bishishe,” “Hirmata Merkato,” “Kochi,” and “Agip” found in Jimma Town. Equal numbers of samples (45 each, totally 360 samples) were collected from the selected markets. The samples were collected, put in plastic bags, properly labeled, and brought to the Medical Parasitology Laboratory of Jimma University, for parasitological analysis.
A portion (200 g) of each fruit and vegetable was washed separately in 500 mL of normal saline for detaching the parasitic stages (ova, larvae, cysts, and oocysts) of helminths and protozoan parasites commonly assumed to be associated with vegetable contamination. After overnight sedimentation of the washing solution, 15 mL of the sediment was then transferred to a centrifuge tube using sieve, to remove undesirable matters. For concentrating the parasitic stages, the tube was centrifuged at 3000 rpm for five minutes [5 ]. After centrifugation, the supernatant was decanted carefully without shaking. Then the sediment was agitated gently by hand for redistributing the parasitic stages. Finally, the sediment was examined under a light microscope using ×10 and ×40 objectives. Modified Zeihl-Neelsen staining technique was also used for identification of oocysts of Cryptosporidium and Cyclospora spp as described elsewhere [11 ].

We considered various ways of supplying a daily, freshly prepared, safe, and palatable portion of food that contained green leafy vegetables, fruit, and milk to women who were living across an urban slum area ∼13 × 13 km. The best solution, after development and pilot testing in a different slum community (Shetanchowki, Mumbai), was snacks that resembled local street foods such as samosas and fritters, which could be filled with the key ingredients, cooked, packaged, and easily transported (13 ).
Treatment snacks contained fresh and dried green leafy vegetables, milk, and dried fruit (Table 1). Green leafy vegetables included spinach, colocasia, amaranth, fenugreek, coriander, shepu, spring onion stalk, and curry leaves. Initially, we used dried green leafy vegetables with the rationale being to provide more micronutrients per unit volume of green leafy vegetables. These vegetables were commercially produced, air-dried at room temperature, and supplied as powders or flakes. However, as the trial progressed, we increased the proportion of fresh leaves purchased from local markets, which improved the palatability without major changes in the nutrient content (Table 2). Dried fruits included figs, dates, raisins, mango, apple, gooseberry, and guava. Milk was included as commercially bought full-fat milk powder. Control snacks were made from low-micronutrient vegetables such as potato, tapioca, and onion, which were purchased from local markets. To avoid monotony for the women, we created 70 treatment and 40 control recipes from these foods of which 8–14 were in use at any time (see Supplemental Table 1 under “Supplemental data” in the online issue). Snacks were made fresh each day in a dedicated study kitchen at the CSSC. Both treatment and control snacks had similar added spices, bindings, and covering ingredients (wheat, rice, or chickpea flour and semolina) and (except for one recipe in each allocation group) were cooked by deep frying in sunflower oil.
We aimed to improve diet quality rather than specific nutrient intakes by raising intakes of green leafy vegetables, fruit, and milk. On average, treatment snacks contained 10–23% of the WHO/FAO recommended Reference Nutrient Intakes for β-carotene, riboflavin, folate, vitamin B-12, calcium, and iron (Table 2) (15 ). Snacks were tested approximately every 6 mo for micronutrient contents (Eclipse Ltd), and microbiological contamination (coliforms and aflatoxin; Intertek Testing Services) with consistently negative results.

The risk flow in Figure 1 represents the effectiveness of launching AI to select policies in an environment. Using an OpenAI-gym library, the system simulates mango allocation during the export period in a customized risk field. This library is a toolkit for reinforcement learning. It includes several benchmark problems that expose standard interfaces and compare algorithm performance (Brockman et al., 2016 ). The simulation system with this library can construct a scene of fruit logistics, such as the fruit loss process (loss at the farm level, loss due to transportation, loss at the wholesale level, loss during storage, loss at the retail level, loss at the consumer level, and loss during processing). In the simulation, the intelligent agent drives the reaction process. Only a small probability of mango loss may occur using the FIFO and LSFO policies, as shown in Joas et al. (2010 (link)).

The proposed smart contract broadcasts LSTM AI model parameters. Discrete mango data owners can upload local model parameters globally, while other people can read them. To design an intelligent technology policy, the developer needs scale-free reinforcement learning to compute the execution timing, whether launching a FIFO or LSFO method when processing a mango export strategy (Kaelbling et al., 1996 (link)). In addition, the forecasting model and the LSTM model, which predicts the mango maturity trend (Bruckner et al., 2013 (link)), supports AI decision-making by sharing its features on the blockchain.
The value of a blockchain lies in its ability to store intelligence. The smart contract aggregation feature allows for sharing AI models without needing middlemen. Spreading machine knowledge through the hive mind platform is likely to solve one of the enormous supply chain normalization and fruit maturity consensus problems. Just as the internet allows websites to spread information, smart contracts allow the broadcasting of model parameters. After data collection by sensors, the AI model has been trained using these datasets. Subsequently, the model predicts the ripeness of the 20% trend in future projections. The prediction uses the trained intelligence to generate a meaningful equation-free model (Kevrekidis and Samaey, 2010 (link)) to measure the relative ripeness momentum. These processes allow authorized users to vote on forecasting model proposals on the blockchain, choosing whether to merge the old and new forecasting models. However, it should be noted that there is a fee for running a smart contract. Every time a smart contract is executed, a fee must be paid to the EVM for execution. This fee is paid to the nodes that help store, compute, execute, and validate smart contracts. EVM is known as the core of Ethereum, demonstrating its importance to the Fantom network (Choi et al., 2018 ; Kaur and Gandhi, 2020 (link)) (layer 1).
Layer 1 is a blockchain architectural term that refers to a network that provides infrastructure or consensus on projects, such as an event-based coffee supply chain (Bager et al., 2022 (link)). A virtual machine (VM) is a computer system with complete hardware functions simulated by software and running in a completely isolated environment. By generating a new virtual image of the existing operating system, the VM performs the same functions as the Windows system; however, it runs independently from a Windows system. As the name suggests, the EVM is Ethereum's VM. Notably, there are no VMs in the Bitcoin blockchain (Nakamoto, 2008 (link)). The primary function of Bitcoin is to store data in a distributed manner, and we can record, verify, store, and replicate transaction data in this network. Ethereum is a “decentralized real-world computer,” and developers can also build DApps on it, implying that Ethereum not only needs to be able to distribute data storage but also needs to run code and conduct consensus communication (Tikhomirov, 2017 (link); Hildenbrandt et al., 2018 (link)). If an account wants to execute a smart contract, the transfer will be completed according to the smart contract, and the relevant execution rules will be recorded in the data to guide the contract's operation. The network nodes execute the smart contract code every time the described transaction occurs through the EVM.

The proposed system contributes to a potential Supply Chain 4.0 for addressing the issues discussed in Section 2. The ISM (Pfohl et al., 2011 (link); Astuti et al., 2014 (link)) transforms the operation flow of the “Anto Wijaya Fruit” company (Widi et al., 2021 (link)) into our design for process flows covering product, finance, and risk, as outlined below.
where Net CE denotes net cost-efficiency,
EV denotes the export value of the mango,
L denotes mango loss (%),
R denotes risk % as a result of the less-than-ideal decision to reduce mango loss,
N denotes the number of sensing devices,
Cost of each sensing device = 200 USD,
x denotes the additional cost per sensing device.
In this case, x = sensor storage cost + implementation cost + installation cost + broadband transmission cost + visualization equipment cost. The details are as follows:
Assuming a total additional cost per sensing device (x) is US $20 for the sensor's life span (10 years), the total upfront cost of purchasing and operating each sensor is US $200 + US $20 = US $ 220. The profit recovered by this AI system is EV^*(L%–R%), where L = 6.91% and R = 0.035%. This could be used as the budget for paying the initial sensor cost. The maximum number of sensors that could be purchased using this profit is listed in Table 1.